TESTING THE LIMITS OF THE FIRST AMENDMENT: HOW ONLINE CIVIL RIGHTS TESTING IS PROTECTED SPEECH ACTIVITY

TESTING THE LIMITS OF THE FIRST AMENDMENT: HOW ONLINE CIVIL RIGHTS TESTING IS PROTECTED SPEECH ACTIVITY

This Note assesses First Amendment freedom of speech claims with regard to online civil rights testing. Transactions that have conventionally occurred in person are now more often completed online, and providers transacting online have been increasingly using algo­rithms that synthe­size users’ data. While these algorithms are helpful tools, they may also be yielding discriminatory results, whether inten­tionally or unintentionally.

In order to test whether such algorithms are discriminating, civil rights testers and researchers have developed various online auditing methods. Two methods in particular, the “sock puppet” audit and the “scraping” audit, have been considered especially viable in de­tecting discrimination. In a sock puppet audit, a tester acts as different bona fide patrons of various demographic backgrounds to test whether an algorithm returns disparate results. In a “scraping” audit, the tester creates bots that act as different individuals and then issue repeated queries to an algorithm and record the various responses received.

While these methods are promising, they currently violate many online providers’ terms of service. Further, due to judicial interpretations of the Computer Fraud and Abuse Act (CFAA) as proscribing violations of a web­site’s terms of service, engaging in these testing methods could give rise to criminal liability.

This Note argues that, in light of precedent related to protected conduct, false speech, investigative journalism, and the tradition of testers in civil rights enforcement, the First Amendment’s protections can and should extend to civil rights testing. Therefore, the CFAA, inso­far as it is applied to civil rights testers engaging in online testing activity, infringes upon the First Amendment by criminalizing constitutionally protected conduct.

Introduction

Access to evidence of discrimination has historically been seen as one of the greatest challenges for civil rights plaintiffs. 1 See Reva Siegel, Why Equal Protection No Longer Protects: The Evolving Forms of Status-Enforcing State Action, 49 Stan. L. Rev. 1111, 1136–37 (1997) (stating that the dis­criminatory purpose standard is one that “plaintiffs will rarely be able to prove”). Gathering such evidence will likely become even more difficult as transactions that traditionally occurred in person—like renting homes, applying for loans, and buying goods—go online. 2 Avnita Lakhani, Introduction to Commercial Transactions in the Virtual World: Issues and Opportunities 3, 19 (Avnita Lakhani ed., 2014). Discrimination has taken on a new face, hiding behind algorithms that crunch big data, 3 See infra section I.A (giving examples of algorithm-based discrimination). which is the data culled from individuals’ online activity. 4 See Steve Lohr, The Age of Big Data, N.Y. Times (Feb. 11, 2012), http://www.nytimes.
com/2012/02/12/sunday-review/big-datas-impact-in-the-world.html?_r=1 (on file with the Columbia Law Review) [hereinafter Lohr, Age of Big Data].
And as discriminatory tactics evolve, so must the ways in which they are rooted out.

To that end, researchers have developed a number of methods to assess whether algorithms behave discriminatorily. 5 See Christian Sandvig et al., Auditing Algorithms: Research Methods for Detecting Discrimination on Internet Platforms 8–16 (2014), http://pdfs.semanticscholar.org/b722/
7cbd34766655dea10d0437ab10df3a127396.pdf [http://perma.cc/W6CK-LHEU] (outlining dif­ferent test designs for auditing algorithms).
Many draw upon the conventional method for uncovering discriminatory practices: the use of testers, individuals who pose as potential consumers, renters, or employees for the sole purpose of testing whether an entity engages in discrimina­tory treatment in violation of civil rights statutes. 6 See Tandy v. City of Wichita, 380 F.3d 1277, 1285–87 (10th Cir. 2004) (defining “testers” and holding that disability testers have standing to bring suit).

Two methods of testing—the “scraping” audit and the “sock puppet” audit—are particularly promising. 7 Sandvig et al., supra note 5, at 12–14. In a “scraping” audit, a researcher uses a bot to impersonate users of various backgrounds; the bot then issues repeated queries to see how an algorithm functions in response and sub­sequently collects the responses it receives. 8 Id. at 12–13. A “sock puppet” audit essen­tially replicates a traditional testing scenario online: Researchers imitate users of various backgrounds with fake accounts or prepro­grammed data profiles to test whether any differential treatment occurs. 9 Id. at 12–14.

These methods have already proven to be effective. Researchers inves­tigating racial discrimination on Airbnb, an online platform that facilitates short-term rentals and homestays, conducted a study in which twenty user profiles were constructed and then used to send rental re­quests to approximately 6,400 Airbnb hosts. 10 See Ray Fisman & Michael Luca, Fixing Discrimination in Online Marketplaces, Harv. Bus. Rev. (Dec. 2016), http://hbr.org/2016/12/fixing-discrimination-in-online-marketplaces [http://perma.cc/268Q-NGJF]. The sole difference among the requests was the requesting user’s name. 11 Id. Half the profiles used names that were more common among white individuals, while the other half used names more common among African American individuals. 12 Id. The commonality of the names was determined by studying birth records. Id. Through this sock puppet audit, researchers discovered that the requesters with black-sounding names were 16% less likely than those with white-sounding names to have their requests accepted. 13 Id. The discrimination was recorded among both cheaper listings and more expensive ones, and the majority of hosts who declined the users with names common among African American individuals had never hosted an African American guest, suggesting the discrimination was prevalent and entrenched, especially among certain hosts. 14 Id.

The issue with these methods is that the Computer Fraud and Abuse Act (CFAA), 15 18 U.S.C. § 1030 (2012). a cybersecurity statute, effectively prohibits their use. Many websites employ terms of service that prohibit scraping 16 See, e.g., Terms of Use, Zillow, http://www.zillow.com/corp/Terms.htm [http://
perma.cc/CT6H-45WT] (last updated Mar. 2016) (“Automated queries (including screen and database scraping, spiders, robots, crawlers and any other automated activity with the pur­pose of obtaining information from the Services) are strictly prohibited on the Services . . . .”).
and the crea­tion of false profiles, 17 See, e.g., United States v. Drew, 259 F.R.D. 449, 461–62 (C.D. Cal. 2009) (addressing a case in which the defendant was charged with a violation of the CFAA after she created a fake MySpace profile, because MySpace mandates that profiles only be created with “truth­ful and accurate” information); Terms of Use, Monster.com, http://inside.monster.com/terms-of-use [http://perma.cc/6MLR-A6BR] (last updated Mar. 2014) (preventing the posting or submission of “any incomplete, false or inaccurate biographical information”). By pre­venting the creation of profiles with false information, these terms of service prohibit sock puppet auditing, which relies on the creation of false profiles. See supra notes 9–14 and accompa­nying text (describing the sock puppet audit methodology). and courts have interpreted the CFAA to penalize violations of a website’s terms of service. 18 See infra note 142 (listing cases in which such an interpretation has been espoused). Thus, the CFAA presents a substan­tial obstacle to researchers, journalists, and civil rights activists hoping to use online civil rights testing methods. A recent case, Sandvig v. Sessions, brought this problem to the forefront. The American Civil Liberties Union (ACLU) filed the case on behalf of researchers and jour­nalists who would like to engage in these methods of online civil rights testing but are chilled from doing so as a result of the CFAA. 19 See infra notes 124–129 (detailing the Sandvig v. Sessions case). The case raises a novel theory: The CFAA is unconstitutional because the test­ing activity it effectively prohibits is protected by the First Amendment. 20 See infra notes 124–129.

This Note assesses these freedom of speech and expression claims with regard to online civil rights testing and argues that, in light of past precedent related to protected conduct, false speech, investigative jour­nalism, and the tradition of testers in civil rights enforce­ment, First Amend­ment protections should extend to civil rights testing. Therefore, applying the CFAA to online civil rights testers infringes upon testers’ First Amend­ment rights. The focus of this Note is limited to the substantive question of whether civil rights testing, and more specifically online testing, is a type of conduct that the First Amendment reaches; it does not dwell upon related issues that may arise over the course of litigation. 21 There is a question of whether the state action requirement, necessary for First Amendment scrutiny to be triggered, has been met in the Sandvig suit. Though this Note touches briefly on the issue, see infra notes 229–230 and accompanying text, a complete analysis of that issue is outside the scope of this Note. The Supreme Court has noted that the state action requirement calls only for an assessment of “whether [state] power has in fact been exercised,” even if that occurs through a court’s application of a tort at common law. N.Y. Times Co. v. Sullivan, 376 U.S. 254, 265 (1964). Furthermore, “[s]tate action might be found in state support or encouragement of private choice; the involvement of police or the courts in enforcing private decisions; . . . or the acknowledgement that when the state has the capacity to act, the absence of state involvement is itself a choice.” Christopher W. Schmidt, The Sit-Ins and the State Action Doctrine, 18 Wm. & Mary Bill Rts. J. 767, 780–81 (2010) (footnotes omitted). However, in more recent years, the Supreme Court has pulled back on the state action doctrine. See Theodore Y. Blumoff, Some Moral Implications of Finding No State Action, 70 Notre Dame L. Rev. 95, 97–100 (1999) (de­scribing how the Court limited the notion of what constitutes state action, especially with the case DeShaney v. Winnebago Cty. Dep’t of Soc. Servs., 489 U.S. 189 (1989)). Thus, though the answer as to whether the state action requirement is satisfied in this case is not fully settled, this Note proceeds on the assumption that state action will be met through prosecution or sanctioning of testers under the CFAA. See also infra notes 231–237 and ac­companying text (discussing the public forum requirement with regard to the CFAA–First Amendment issue).

Part I describes specific instances of algorithm-based discrimination that have been discovered and outlines the history of civil rights testing and First Amendment doctrine to understand the CFAA–First Amendment problem. Part II analyzes the act of online civil rights testing under the doctrinal frameworks for false speech and conduct incidental to speech, each of which could provide a basis for First Amendment protection of testing conduct; it concludes that the CFAA, as applied to testers, and the First Amendment are in tension because online civil rights testing can be protected under both frameworks. Finally, Part III argues that First Amend­ment protections should be pur­sued through litigation even though the CFAA could be amended to allow for a tester exception. Whether the CFAA is amended, basing citizens’ right to test for online discrimination in con­stitutional law will offer civil rights advocates, journalists, and researchers firmer, more lasting protection in accessing information about, and working against, discrimina­tion in the age of big data.

I. Algorithm-based Discrimination, Testers, and the First Amendment

This Part contextualizes the filing of Sandvig v. Sessions. Namely, it high­lights the emerging and widespread issue of algorithm-based discrimination, the history and importance of civil rights testing, and the doctrinal background that will inform an assessment of whether First Amendment protection covers the conduct of civil rights testing. Section I.A.1 discusses the recent phenomenon of online algorithms and their discriminatory impact to date. Section I.A.2 turns to civil rights testers, describing their traditional function and how testing has been retooled to address algorithm-based discrimination. Section I.B then summarizes the legal frameworks for two different lines of First Amendment case law, each of which could protect these new, online methods of civil rights testing.

A. Algorithm-based Discrimination and the Tool of Testers

1. Big Data and the Advent of Online Algorithm-based Discrimination. — Though the concept of big data is relatively new, it has become promi­nent in discussions of the internet and technology, especially in relation to online privacy. As used in this Note, big data is the “massive quantities of information produced by and about people, things, and their inter­actions,” culled from the “digital traces left by people” through their online activities. 22 Danah Boyd & Kate Crawford, Critical Questions for Big Data, 15 Info. Comm. & Soc’y 662, 663 (2012). Others have described big data as “the data of the Web—online searches, posts and messages.” Lohr, Age of Big Data, supra note 4. As these data become increasingly available, tools for leveraging them develop as well. 23 See Lohr, Age of Big Data, supra note 4 (“[T]he computer tools for gleaning knowledge and insights from the Internet era’s vast trove of unstructured data are fast gaining ground.”). Algorithms are one such type of tool. A number of companies and groups are turning to algorithms to synthesize large amounts of data into valuable predictions. 24 See, e.g., id. (“Police departments across the country, led by New York’s, use com­puterized mapping and analysis of variables like historical arrest patterns, paydays, sporting events, rainfall and holidays to try to predict likely crime ‘hot spots’ and deploy officers there in advance.”). This can often be useful to consumers: For example, algorithms can be used to predict what books a consumer may like based on her past responses to suggestions. 25 Andrew McAfee & Erik Brynjolfsson, Big Data: The Management Revolution, Harv. Bus. Rev. (Oct. 2012), http://hbr.org/2012/10/big-data-the-management-revolution [http://
perma.cc/YS6P-2MUA] (“Before long, [Amazon] developed algorithms to predict what books individual customers would like to read next—algorithms that performed better every time the customer responded to or ignored a recommendation.”).
However, algorithms can also be keyed to search for certain characteristics and thereby enable discrimination. 26 See Solon Barocas & Andrew D. Selbst, Big Data’s Disparate Impact, 104 Calif. L. Rev. 671, 692–93 (2016) (explaining the concept of “masking,” in which “decision makers with prejudicial views can mask their intentions by exploiting [algorithmic] mechanisms”).

Algorithm-based discrimination, as used in this Note, is defined as the unequal results generated by the explicit or implicit biases built into algorithms. Explicit bias, or disparate treatment, occurs when algorithms are created or used to purposely discriminate against individuals. For instance, one linguist has developed software that can both predict an individual’s ethnicity based on the individual’s full name, address, and ZIP code and identify 158 ethnicities based on predictive algorithms that use different ethnic groups’ name patterns and demographic data. 27 See Perry Garfinkel, A Linguist Who Cracks the Code in Names to Predict Ethnicity, N.Y. Times (Oct. 15, 2016), http://www.nytimes.com/2016/10/16/jobs/a-linguist-who-cracks-
the-code-in-names-to-predict-ethnicity.html (on file with the Columbia Law Review).
Though this software’s contracts specify permissible uses to prevent any possible “suspicious motivations,” 28 Id. not every company that develops or uses pre­dictive algorithms may safeguard against discriminatory usage. 29 For some examples on how algorithms have been used discriminatorily, see Claire Cain Miller, When Algorithms Discriminate, N.Y. Times: The Upshot (July 9, 2015), http://
www.nytimes.com/2015/07/10/upshot/when-algorithms-discriminate.html (on file with the Columbia Law Review); see also Council of Econ. Advisers, Big Data and Differential Pricing 16–17 (2015), http://obamawhitehouse.archives.gov/sites/default/files/whitehouse_files/
docs/Big_Data_Report_Nonembargo_v2.pdf [http://perma.cc/LH5E-5U3B] (“In principle, big data could lead to disparate impacts by providing sellers with more variables to choose from, some of which will be correlated with membership in a protected class.”).

In the absence of this kind of disparate treatment, or purposeful discrimination, algorithms can nonetheless lead to discrimination by causing a disparate impact against certain groups. 30 In fact, scholars seem to find it more likely that algorithms result in disparate impact than disparate treatment. See Barocas & Selbst, supra note 26, at 693 (“When it comes to data mining, unintentional discrimination is the more pressing concern because it is likely to be far more common and easier to overlook.”). For instance, the use of proxies can result in discrimination when certain information corre­sponds with membership within a particular protected class. 31 See id. at 691–92. Disparate impact can also occur when an algorithm contains implicit bias, such as when it is based on examples 32 These examples are also known as “training data.” Id. at 680. Training data are the data that are used to model algorithms and teach them. See Shan Suthaharan, Machine Learning Models and Algorithms for Big Data Classification 7 (2016). in which prejudice played a role in the decisionmaking. 33 Barocas & Selbst, supra note 26, at 680. For instance, St. George’s Hospital in the United Kingdom historically discriminated against racial minorities and women in making medical school admissions decisions. Id. at 682. When much later the Hospital used a pro­gram to sort medical school applicants that relied on data culled from these discrimina­tory decisions, the Hospital unknowingly perpetuated those same past biases. Id. While disparate impact is not always legally actionable, 34 Disparate impact causes of action are cognizable under some statutes, such as Title VII of the Civil Rights Act and the Fair Housing Act, see Tex. Dep’t of Hous. & Cmty. Affairs v. Inclusive Cmtys. Project, Inc., 135 S. Ct. 2507, 2525 (2015), but a theory of dis­crimination premised on disparate impact cannot support a claim under the Equal Protection Clause, Washington v. Davis, 426 U.S. 229, 240 (1976). testers who uncover disparate impact can still use their findings to seek change through nonlegal avenues.

Researchers have found or suspected that algorithms are working discriminatorily in a number of contexts. For instance, advertisers using algorithms may target groups in prejudicial ways by presenting different advertisements to different individuals based on their sex or race. 35 This is also known as targeted advertising, and advertisers often turn to companies that harvest and sell demographic data for help in targeting potential consumers. See Joseph Turow, The Daily You: How the New Advertising Industry Is Defining Your Identity and Your Worth 95–97 (2012). Some of these companies even include telecom industry giants like Verizon and AT&T, though the Federal Communications Commission has promulgated a rule that requires these providers to get consumer permission, thereby limiting their ability to harvest and sell data for targeted advertising. See Sapna Maheshwari & Cecilia Kang, Telecoms’ Ambitions on Targeted Ads Seen Curbed by F.C.C.’s New Privacy Rules, N.Y. Times (Oct. 28, 2016), http://www.nytimes.com/2016/10/29/business/media/telecoms-
ambitions-on-targeted-ads-seen-curbed-by-fccs-new-privacy-rules.html?_r=0 (on file with the Columbia Law Review).
This was demonstrated in two recent examples: Research showed that Google displayed an advertisement for high-income jobs more often to men than to women, 36 Miller, supra note 29. This discovery was made through the use of false profiles, otherwise known as sock puppet auditing. See Amit Datta, Michael Carl Tschantz & Anupam Datta, Automated Experiments on Ad Privacy Settings: A Tale of Opacity, Choice, and Discrimination, 1 Proc. on Privacy Enhancing Techs. 92, 92 (2015). and advertisements for arrest records were significantly more likely to show up on searches for historically black fraternities or for names considered “distinctively black.” 37 Miller, supra note 29.

Algorithms may also lead to discriminatory outcomes by differenti­ating the prices of goods 38 Differential pricing of goods is the act of charging different customers varying prices for the same products or services. Council of Econ. Advisers, supra note 29, at 4. Differential pricing may also be called “price discrimination.” Id. based on protected characteristics like race or ethnicity. 39 Id. at 16 (“Given hundreds of variables to choose from, it is easy to imagine that statistical models could be used to hide more explicit forms of discrimination by gener­ating customer segments that are closely correlated with race, gender, ethnicity, or religion.”). The use of proxies is problematic not only because it can camouflage intentional discrim­ination but also because it can cause a disparate impact. One example of such a disparate impact is Amazon’s same-day Prime servicing. In a number of cities, such as Atlanta and Chicago, “black citizens are about half as likely to live in neighborhoods with access to Amazon same-day delivery as white residents,” and in Boston, three ZIP codes that “en­compass[ed] the primarily black neighborhood of Roxbury [were] excluded from same-day ser­vice.” David Ingold & Spencer Soper, Amazon Doesn’t Consider the Race of Its Customers. Should It?, Bloomberg (Apr. 21, 2016), http://www.bloomberg.com/graphics/2016-amazon-same-day/ [http://perma.cc/ZT6H-5FQ4]. Although Amazon purportedly does not take race into account, it does take into account the number of residents who have purchased the $99 annual Prime membership when it makes service-area decisions. Id. If Prime member­ship correlates to income, areas with more Prime members would also be wealthier areas. Thus, where there is a wealth disparity between various races, Amazon’s differential servicing would surely be a case of disparate impact. And if a company such as Amazon knew of these correlations, it could engage in intentional discrimination by using neutral proxies, such as Prime membership. See also Barocas & Selbst, supra note 26 (discussing “masking” dis­criminatory intent with algorithms). For example, The Princeton Review, a test-preparation-services company, used a differential pricing model that showed different prices to individuals across various geographic areas. 40 Lauren Kirchner, When Discrimination Is Baked into Algorithms, Atlantic (Sept. 6, 2015), http://www.theatlantic.com/business/archive/2015/09/discrimination-algorithms-
disparate-impact/403969 [http://perma.cc/WFJ8-6XX4].
This model, however, mapped onto racial demographics: Customers in geographic areas with higher Asian populations were charged more for online SAT tutoring. 41 Id. This pricing scheme was likely not based on income, since lower-income neighborhoods with large Asian populations were still shown higher prices. 42 See Julia Angwin & Jeff Larson, The Tiger Mom Tax: Asians Are Nearly Twice as Likely to Get a Higher Price from Princeton Review, ProPublica (Sept. 1, 2015), http://
www.propublica.org/article/asians-nearly-twice-as-likely-to-get-higher-price-from-princeton-review [http://perma.cc/YCP2-Z4CJ]. It is further worth noting that a differential pricing scheme based on a geographic area’s average income in which higher prices are displayed to indi­viduals living in ZIP codes with a lower weighted average income, a side effect of the scheme used by Staples.com, see Jennifer Valentino-DeVries, Jeremy Singer-Vine & Ashkan Soltani, Websites Vary Prices, Deals Based on Users’ Information, Wall St. J. (Dec. 24, 2012), http://www.wsj.com/articles/SB10001424127887323777204578189391813881534 (on file with the Columbia Law Review), can give rise to disparate impact against minorities who live less affluently than their neighbors. Alternatively, entities that wish to engage in intentional discrimination based on race might use income as a proxy and thus cause disparate treat­ment against minorities.
The ProPublica article that initially broke this story referred to the price differential as a “tiger mom tax” 43 Angwin & Larson, supra note 42. —referencing a familiar stereotype of Asian American parents 44 For background on the “tiger mom” stereotype, see generally Linda P. Juang, Desiree Baolin Qin & Irene J. K. Park, Deconstructing the Myth of the “Tiger Mother”: An Introduction to the Special Issue on Tiger Parenting, Asian-Heritage Families, and Child/
Adolescent Well-being, 4 Asian Am. J. Psychol. 1, 1–3 (2013).
and raising the uncomfortable idea that The Princeton Review sought to profit based on this stereotype.

Employment 45 Solon Barocas and Andrew Selbst have explored the possible interactions between big data and Title VII by hypothesizing the ways in which employment discrimination might manifest itself through big data. See Barocas & Selbst, supra note 26, at 683. and housing are additional areas in which algorithms can lead to discriminatory effects. In the recruiting context, algorithms could cause disparate impact if they are keyed to look for certain character­istics, such as academic credentials, and the schools they are programmed to select for have a stark lack of diversity. 46 Id. at 689. Alternatively, algorithm-based “talent-matching” tools could cause purposeful discrimination if the algo­rithms they use to match employees with employers take into account employers’ prejudicial preferences. 47 Id. at 692. Similarly, since big data is also used to facilitate housing transactions, 48 Miller, supra note 29. these data, which could come from both public sources like the Census Bureau and smartphone apps that draw data from users’ recorded locations, can lead to purposeful discrimination by “racially steer[ing] consumers—to rent or buy in Black neighbor­hoods or to stay away from them.” 49 Alex Rosenblat et al., Data & Civil Rights: Housing Primer 5 (2014), http://www.
datacivilrights.org/pubs/2014-1030/Housing.pdf [http://perma.cc/9D52-JMU8]; see also FTC, Big Data: A Tool for Inclusion or Exclusions? Understanding the Issues 25 (2016), http://
www.ftc.gov/system/files/documents/reports/big-data-tool-inclusion-or-exclusion-understanding-
issues/160106big-data-rpt.pdf [http://perma.cc/37SE-SXB8] (noting that “[n]umerous re­searchers and commenters discuss how big data could be used in the future to the dis­advantage of low-income and underserved communities and adversely affect consumers on the basis of legally protected characteristics” in areas like housing).

The financial sector may also engage in algorithm-based discrimina­tion when assessing applications for lending services. 50 Rosenblat et al., supra note 49, at 4 (“[A]lgorithmic analysis may differentially evaluate creditworthiness according to race because of the confounding nature of neigh­borhood data.”). For instance, new software for bankers uses big data to interpret signals about behavior (such as whether customers use proper capitalization) to aid in assessing creditworthiness. 51 Steve Lohr, Banking Start-Ups Adopt New Tools for Lending, N.Y. Times (Jan. 18, 2015), http://www.nytimes.com/2015/01/19/technology/banking-start-ups-adopt-new-tools-for-
lending.html?_r=0 (on file with the Columbia Law Review).
Such software could discriminate against certain racial or ethnic groups, even if not programmed to do so, if a particular behavior that is taken into account is more prevalent among certain protected classes. 52 Id. In addition to this example, lending software recently patented by Facebook, which uses the credit ratings of an individual’s social network “friends” in assessing that individual’s risk, also raises questions of discrimination. See Jonathan Zim, The Use of Social Data Raises Issues for Consumer Lending, U. Miami Bus. L. Rev.: UMBLR Insights (Apr. 28, 2016), http://business-law-review.law.miami.edu/social-data-raises-issues-consumer-lending/ [http://perma.cc/R86M-5NDV]. The National Consumer Law Center voiced concerns about how this lending program might result in racial discrimination, because the credit scores of African Americans, which tend to be lower, largely reflect twentieth-century biases, biases that would then be perpetuated by the lending software. Id. These worries may be reasonable not only based on past biases but also because of how risk assessments have used algorithms to generate discriminatory outcomes as well. ProPublica found that an algo­rithm used to conduct criminal risk assessments gave biased ratings. The Hidden Discrimination in Criminal Risk-Assessment Scores, NPR (May 24, 2016), http://www.npr.org/2016/05/24/
479349654/the-hidden-discrimination-in-criminal-risk-assessment-scores (on file with the Columbia Law Review) (“So black defendants are twice as likely to be rated high risk incor­rectly, mean­ing they did not go on to reoffend. And white defendants are twice as likely to be rated incorrectly as low risk and yet go on to reoffend.”).

The above instances show the potential for big data to be used in per­petuating civil rights violations. 53 Disparate impact and disparate treatment are both actionable under the Equal Credit Opportunity Act. See Consumer Fin. Prot. Bureau, Equal Credit Opportunity Act (ECOA) 1 (2013), http://files.consumerfinance.gov/f/201306_cfpb_laws-and-regulations_
ecoa-combined-june-2013.pdf [http://perma.cc/X9CH-HUXU]. Disparate treatment is action­­able under the public accommodations section of the Civil Rights Act of 1965. See 42 U.S.C. § 2000a (2012); see also supra note 34 (explaining that statutory claims of dis­crimination in employment and housing can be based on disparate impact, but Equal Protection Clause claims must be premised on purposeful discrimination).
Unanticipated uses of big data, espe­cially as these online tools become more sophisticated, amplify the threat of discrimination, and plaintiffs have begun to bring antidiscrimi­nation suits as a result. 54 Recently, for example, plaintiffs have sued companies such as Amazon and T-Mobile for age discrimination; these companies used big data generated by Facebook to determine which Facebook users would be shown their employment advertisements, and they then limited their advertisements to younger users. See generally Complaint, Bradley v. T-Mobile US, Inc., No. 5:17-cv-07232 (N.D. Cal. filed Dec. 20, 2017). In order to monitor this new form of civil rights viola­tions, advocates will have to update a traditional tactic for rooting out dis­crimination: testers.

2. The Tester Tradition. — The use of testers, individuals whose only aim is to test whether a particular entity is engaging in unlawful discrimi­nation by posing as a potential patron, employee, or tenant, 55 See Tandy v. City of Wichita, 380 F.3d 1277, 1285 (10th Cir. 2004) (defining “testers”). has become a vital tool in detecting discrimination in areas such as housing 56 See Stephen E. Haydon, Comment, A Measure of Our Progress: Testing for Race Discrimination in Public Accommodations, 44 UCLA L. Rev. 1207, 1216 (1997). and employment. 57 See Alex Young K. Oh, Note, Using Employment Testers to Detect Discrimination: An Ethical and Legal Analysis, 7 Geo. J. Legal Ethics 473, 475 (1993) (noting that dis­covery of discrimination would be impossible without testers). The Supreme Court recognized the role of testers in civil rights suits in Havens Realty Corp. v. Coleman, 58 455 U.S. 363 (1982). holding that testers have standing to bring claims under section 804(d) of the Fair Housing Act. 59 Id. at 374.

Following Havens Realty, a number of courts of appeals recognized tester standing under other civil rights statutes. 60 See, e.g., Colo. Cross-Disability Coal. v. Abercrombie & Fitch Co., 765 F.3d 1205, 1211 (10th Cir. 2014) (extending tester standing under Title III of the Americans with Disabilities Act (ADA)); Houston v. Marod Supermarkets, 733 F.3d 1323, 1332 (11th Cir. 2013) (same); Tandy, 380 F.3d at 1287 (extending tester standing under Title II of the ADA); Smith v. Pac. Props. & Dev. Corp., 358 F.3d 1097, 1104 (9th Cir. 2004) (extending tester standing under the Fair Housing Amendments Act); Kyles v. J.K. Guardian Sec. Servs., 222 F.3d 289, 300 (7th Cir. 2000) (extending tester standing under Title VII of the Civil Rights Act of 1964). In doing so, these courts highlighted the importance of testers in achieving the antidiscrimination ends Congress intended with these statutes. The Seventh Circuit, for in­stance, noted that testers function akin to private attorneys general, and “because proof of discrimination is often quite difficult to muster . . . testers pro­vide evidence that, we have recognized, ‘is frequently valuable, if not indispensable.’” 61 Kyles, 222 F.3d at 299 (quoting Richardson v. Howard, 712 F.2d 319, 321 (7th Cir. 1983)). Similarly, the Ninth Circuit acknowledged that “[t]esters have played a long and important role in fair housing enforcement.” 62 Smith, 358 F.3d at 1102. As these appellate courts have firmly recognized, testers’ importance comes from their truth-revealing power—that is, their ability to uncover discrim­inatory practices not readily apparent. 63 It is worth noting that testers’ importance is not limited to their ability to bring suit. Oftentimes, the discrimination detected by testers is used to raise awareness and seek reform, rather than to litigate. See, e.g., Our Work: Research, Rest. Opportunities Ctr. United, http://rocunited.org/our-work/ [http://perma.cc/45NC-S7VW] (last visited Feb. 11, 2018) (“ROC’s primary research includes over 5,000 surveys of people who work in res­taurants, hundreds of in-depth interviews with employers and employees, and matched pair audit testing.”). The information that comes from testers can be used in reports and other informational material that highlights discriminatory action in an effort to use nonlegal avenues for change. For an example of such a report derived from using civil rights testers, see generally Rest. Opportunities Ctr. United, The Great Service Divide (2014), http://
rocunited.org/wp-content/uploads/2014/10/REPORT_The-Great-Service-Divide2.pdf [http://
perma.cc/GWR2-DBKL].

For the task of uncovering discrimination in the virtual realm, researchers have adopted and adapted the tool of testers. The two methods that form the focus of this Note are the “scraping” audit 64 In this type of audit, researchers create “bots,” automated programs that are instructed to act like a number of users with demographic data. See Sandvig et al., supra note 5, at 12; see also Complaint for Declaratory & Injunctive Relief at 23, Sandvig v. Sessions, No. 1:16-cv-01368 (D.D.C. June 29, 2016), 2016 WL 3536779 [hereinafter Complaint]. The bots then record the treatment they receive. Id. If a housing site is the subject of the civil rights testing, for example, the bots record the properties shown available to them by “scraping” the data. Id. “Scraping” is the act of automated recording of informa­tion. Id. at 18. and the “sock puppet” audit, 65 The “sock puppet” audit is the method most similar to a traditional testing scenario. In lieu of human actors who are hired to pose as potential customers, tenants, or em­ployees, a human researcher acts as the potential patron by posing as a bona fide website user. Sandvig et al., supra note 5, at 13. The researcher can do this by one of two methods. First, the researcher can create fake user accounts, each of which displays differing demo­graphic data points. Id. Alternatively, researchers can falsify their big data by importing programmatically constructed data so that the data, when analyzed by a site’s algorithms, would portray the researchers as embodying various demographic traits. Id. Like a tradi­tional tester, an online tester pretends to be a bona fide patron, but unlike a traditional tester, online testers misrepresent their demographic information by inventing false data or creating artificial user accounts. both of which can involve impersonating an authentic online patron through the falsification of big data, whether done by researcher-created bots or the researchers themselves. While these methods of testing have been considered the most fruitful for rooting out discrimination, 66 See id. at 12–14. researchers have been hesitant to fully en­dorse the practices due to the threat of violating the Computer Fraud and Abuse Act. 67 Id. at 12–13. As will be discussed in Part II, the CFAA prohibits any act in contravention of a website’s terms of service, and many websites have banned the sock puppet and scraping methods in their terms. 68 See infra section II.A. In order to show how this interpretation of the CFAA conflicts with the First Amendment, this Note next turns to a review of the First Amendment doctrine.

B. First Amendment Doctrine

The First Amendment protects, among other things, “freedom of speech, or of the press,” 69 U.S. Const. amend. I. liberties considered to be of the most precious and fundamental nature in the United States. 70 See William O. Douglas, The One Un-American Act, Nieman Reports, Jan. 1953, at 20, 20 (“Restriction of free thought and free speech is the most dangerous of all subver­sions. It is the one un-American act that could most easily defeat us.”); see also Near v. Minnesota ex rel. Olson, 283 U.S. 697, 707 (1931) (calling freedom of press and speech an “essential personal liberty”). While the fifteen words that create the First Amendment’s speech and press protections seem relatively straightforward, First Amendment doctrine is in a confusing and unclear state, partly because there can be many conflicting rules within the case law. 71 See Robert Post, Reconciling Theory and Doctrine in First Amendment Jurisprudence, 88 Calif. L. Rev. 2353, 2374 (2000). Additionally, many activities protected under the First Amendment are not explicitly written into the Constitution. 72 For instance, freedom of expression is not within the text of the First Amendment, yet the Court has found that the First Amendment extends to expressive conduct. See Texas v. Johnson, 491 U.S. 397, 406 (1989) (extending First Amendment protection to desecration of the American flag as expressive conduct). Another example is campaign finance protec­tions, which derive from the First Amendment freedoms of expression and association despite not being specifically mentioned in the constitutional text. See Buckley v. Valeo, 424 U.S. 1, 21–24 (1976) (discussing how contribution and expenditure limitations for cam­­paign financing implicate First Amendment interests due to their impact on political expression and association). With this in mind, this section will delineate some of the relevant strains of First Amendment jurisprudence and set forth the various frameworks used. Because civil rights testing is conduct, rather than a purely verbal act, this section will look to First Amendment protections for conduct incidental to speech. 73 It is possible, but unlikely, to ground First Amendment protection of online civil rights testers in the line of cases protecting expressive conduct. The Court has noted that First Amendment protection “does not end at the spoken or written word.” Johnson, 491 U.S. at 404. Whether a particular form of nonverbal communication is within the purview of the First Amendment is determined by the Spence test: For conduct to be protected, it must be “sufficiently imbued with elements of communication.” Spence v. Washington, 418 U.S. 405, 409 (1974) (granting First Amendment protection to an exhi­bition of an American flag bearing a peace sign made out of tape in violation of a state statute). Subsequent case law has interpreted the Spence framework to be a two-part test, looking to both the communi­cator’s intent and the likelihood of an audience interpreting the message. See, e.g., Johnson, 491 U.S. at 404 (quoting Spence, 418 U.S. at 410–11).
Notably, the Court has not applied the Spence test in all cases. See, e.g., R.A.V. v. City of St. Paul, 505 U.S. 377, 398 (1992) (applying First Amendment protection to the conduct of burning a cross in someone’s front yard in violation of a law addressing hate crimes without analyzing why the conduct is protected speech); see also Alan K. Chen, Instrumental Music and the First Amendment, 66 Hastings L.J. 381, 389–90 (2015) (“[T]he Court has not rigidly adhered to the Spence test.”). The test has also faced scholarly criticism. See Robert Post, Recuperating First Amendment Doctrine, 47 Stan. L. Rev. 1249, 1252 (1995) [hereinafter Post, Recuperating] (arguing that the Spence test is “transparently and manifestly false” because “it locates the essence of constitutionally protected speech exclusively in an abstract triadic relationship among a speaker’s intent, a specific message, and an audience’s poten­tial reception of that message” when social context is also important). Thus, under a different framework, online civil rights testing could be protected as expressive conduct. However, the Court has not explicitly abrogated the Spence test, and therefore, it remains the guiding doctrine for cases involving nonverbal communications despite its checkered use. Id. (“The Spence test thus appears to have enjoyed the normal life of a rela­tively minor First Amendment doctrine.”).
Given the Spence test’s focus on the intended audience’s interpretation, see id. at 1253, online civil rights testing likely cannot meet the test because it would fail the test’s audience-perception prong. Online testing as an activity does not typically occur with an audience; it involves only a researcher and a computer. See Sandvig et al., supra note 5, at 12–13 (describing methods of online audit testing). While publication of the testing results is an act geared toward an intended audience, it is the testing itself that would need to be expressive conduct in order to be protected, and whether such testing is expressive is questionable. Cf. Snavely v. Arnold, No. 1:08-cv-2165, 2009 WL 1743737, at *3 (M.D. Pa. June 18, 2009) (ruling on a 12(b)(6) motion and finding that investigating an unsolved murder is not inherently expressive conduct). But cf. Kerr v. City & Cty. of S.F., No. C 10-5733, 2012 WL 3877752, at *10 (N.D. Cal. Sept. 6, 2012) (finding that a reasonable fact finder could interpret the act of requesting public records on a particular fund to “convey a message that [the] Fund was being managed and used improperly”); Seth F. Kreimer, Pervasive Image Capture and the First Amendment: Memory, Discourse, and the Right to Record, 159 U. Pa. L. Rev. 335, 377 (2011) (“We would recognize police seizure of, or prosecution for, drafts of letters or manuscripts as an interference with freedom of expres­sion, even if the seizure occurred before the writer had decided to send or publish them, though no designated ‘audience’ had been deprived of their content.”).
However, since a case can be made that testing also involves a verbal act in the form of misrepresentation, this section will discuss false speech jurisprudence as well. 74 Though lying is technically conduct, its suppression automatically triggers First Amendment scrutiny because it is a speech act. See United States v. Alvarez, 567 U.S. 709, 715 (2012) (considering a statute that prohibited lying about military honors to be a content-based speech regulation).

2. Conduct Incidental to Free Speech. — One line of First Amendment jurisprudence focuses on conduct incidental to, or in preparation for, speech. 75 See Citizens United v. FEC, 558 U.S. 310, 336 (2010) (“Laws enacted to control or suppress speech may operate at different points in the speech process.”). This type of conduct has also been called newsgathering or information gathering in some cases, 76 Although “newsgathering” as a term may suggest that these rights are reserved for the press only, courts have rejected special status rights for only the press. See Pell v. Procunier, 417 U.S. 817, 831 (1974) (noting that the press does not get access to certain privileges that members of the general public do not have the benefit of); see also Glik v. Cuniffe, 655 F.3d 78, 84 (1st Cir. 2011) (explaining why newsgathering protections do not turn on “professional credentials or status”). because journalists and activists in these cases seek access to information for the purpose of subsequently engaging in speech. Under the doctrinal strand of conduct incidental to speech, the Court has protected the means of various kinds of speech, whether those are monetary contributions to political campaigns 77 See Buckley, 424 U.S. at 19 (explaining that restricting campaign financing “necessarily reduces the quantity of expression by restricting the number of issues discussed, the depth of their exploration, and the size of the audience reached”), superseded by statute, Bipartisan Campaign Reform Act of 2002, Pub. L. No. 107-155, 116 Stat. 81, as recognized in McConnell v. FEC, 540 U.S. 93 (2003). or access to trials. 78 Richmond Newspapers, Inc. v. Virginia, 448 U.S. 555, 575 (1980) (“In guaran­teeing freedoms such as those of speech and press, the First Amendment can be read as protecting the right of everyone to attend trials so as to give meaning to those explicit guarantees.”).

One aspect of the doctrine is focused on the right of access to gov­ernmental affairs. This line of cases sprang from Richmond Newspapers, Inc. v. Virginia, a case that found a First Amendment right to attend crim­inal trials. 79 Id. at 580. In coming to its decision, the majority focused on the necessity of access in order to write about trials 80 Id. at 576–77. and explained that the First Amendment should be interpreted broadly. 81 Id. at 576 (“For the First Amendment does not speak equivocally. . . . It must be taken as a command of the broadest scope that explicit language, read in the context of a liberty-loving society, will allow.” (internal quotation marks omitted) (quoting Bridges v. California, 314 U.S. 252, 263 (1941))). Subsequently, the Court created a two-prong test to determine whether a First Amendment “right of access” exists, looking first to whether the press or public historically had access to the process, and second to whether public access plays a significant role in the process’s function. 82 Press-Enter. Co. v. Superior Court, 478 U.S. 1, 8–9 (1986) (applying this test in determining whether there is a First Amendment right to access polling places). This test has also been applied by courts of appeals to governmental proceedings other than trials. 83 See PG Publ’g Co. v. Aichele, 705 F.3d 91, 104 (3d Cir. 2013).

More generally, a right to gather information was referenced in Branzburg v. Hayes. 84 408 U.S. 665, 681 (1972) (“[W]ithout some protection for seeking out the news, freedom of the press could be eviscerated.”); see also Jane Bambauer, Is Data Speech?, 66 Stan. L. Rev. 57, 86 (2014) [hereinafter Bambauer, Is Data Speech?] (“[T]he First Amendment protects the right to gather information in some fashion.”). While the Court acknowledged the existence of such a right, it did not flesh out the contours of the right nor what it covered. 85 Branzburg, 408 U.S. at 681. However, it is clear that there are limits: Earlier, the Court had stated that a right to information was not unlimited because “[t]here are few restric­tions on action which could not be clothed by ingenious argu­ment in the garb of decreased data flow.” 86 Zemel v. Rusk, 381 U.S. 1, 16–17 (1965). In that case, the Court held that the Secretary of State’s refusal to validate passports for Cuba did not violate the First Amendment even if it interfered with citizens’ ability to see the effects of the government’s foreign policy. Id.

In determining where the line for protection can be drawn, Professors Alan Chen and Justin Marceau argue every action necessary to speech falls somewhere along a spectrum of activity. 87 Justin Marceau & Alan K. Chen, Free Speech and Democracy in the Video Age, 116 Colum. L. Rev. 991, 1019 (2016) [hereinafter Marceau & Chen, Free Speech]. On one end lie “the most basic elements of conduct that are necessary to engage in communica­tion.” 88 Id. Many actions on this end will not qualify for First Amendment protection because they will be “too attenuated from the actual expressive activity.” 89 Id. An example of this would be the purchase of gasoline to drive to a rally. 90 Id. On the other end lies “the directly communicative element of the expressive process”; an example of this would be the exhibition of videos or art. 91 Id.

Even if, however, conduct falls closer to the “directly communicative element” end of this spectrum and directly implicates the First Amendment, it may also implicate privacy concerns that are crystallized in tort and criminal laws. 92 See Associated Press v. NLRB, 301 U.S. 103, 132–33 (1937) (“The publisher of a newspaper has no special immunity from the application of general laws. He has no special privilege to invade the rights and liberties of others.”). Thus, courts have still held investigative reporters liable for invasion of privacy or other torts they committed in the course of newsgathering. For instance, in Dietemann v. Time, Inc., the Ninth Circuit ruled on a case that involved undercover Life Magazine reporters who posed as patients and took recordings and photographs during their medical visit in order to expose the plaintiff practicing medicine without a license. 93 449 F.2d 245, 245–46 (9th Cir. 1971). This case also involved misrepresentations, because the reporters used the name of someone with whom the plaintiff was familiar to gain entrance. Id. Misrepresentations will be more fully discussed below. See infra section I.B.2. In deciding whether the reporters violated California privacy law, the Ninth Circuit found that when one invites another into their home, they “take[] a risk that the visitor may not be what [they] seem[]” and that this visitor will retell what was heard and seen upon leaving. 94 Dietemann, 449 F.2d at 249. However, one does not take the risk of being photographed or recorded, and the First Amendment does not immunize journalists from liability for this kind of violation. 95 Id. (“The First Amendment has never been construed to accord newsmen immunity from torts or crimes committed during the course of newsgathering.”).

Somewhat contrary to Dietemann, the Seventh Circuit, in Desnick v. American Broadcasting Cos., held that use of test patients, sent with concealed cameras to eye doctors’ offices to see whether improper diagnoses occurred, did not give rise to claims for trespass, invasions of privacy, or wiretapping under Illinois law. 96 44 F.3d 1345, 1352–53 (7th Cir. 1995). Note that the Seventh Circuit did not disagree with the proposition that the First Amendment does not protect against tort lia­bility. See id. at 1351 (“[T]here is no journalists’ privilege to trespass.”). The case distinguished Dietemann by claiming that here, the patients entered offices open to anyone seeking treatment and taped physicians engaging in professional, rather than personal, communica­tions, whereas Dietemann involved a much more private context. 97 Id. at 1352–53; see also Bambauer, Is Data Speech?, supra note 84, at 86 (“Given the long-held expectations that the home provides sanctum, Mr. Dietemann may have won his case despite the First Amendment arguments.” (footnote omitted)). The Seventh Circuit also analogized to testers in the civil rights context in deciding that no trespass occurred, explaining that civil rights testers who enter property without disclosing their true intentions do not inter­fere with ownership or possession of land. 98 Desnick, 44 F.3d at 1353. Courts of appeals that have addressed similar cases have agreed with Desnick, 99 See, e.g., Med. Lab. Mgmt. Consultants v. Am. Broad. Cos., 306 F.3d 806, 818 (9th Cir. 2002); Food Lion, Inc. v. Capital Cities/ABC, Inc., 194 F.3d 505, 517–18 (4th Cir. 1999). though scholars disagree about whether information gathering should be protected. 100 For instance, one scholar, Professor Jane Bambauer, has gone so far as to argue that data collection processes should be protected by a First Amendment right to create knowledge. Bambauer, Is Data Speech?, supra note 84, at 60. Professor Bambauer argues that since at least some facts have been given First Amendment protection, see Va. State Bd. of Pharmacy v. Va. Citizens Consumer Council, 425 U.S. 748, 770 (1976) (holding that infor­mation on prescription medication prices is protected by the First Amendment), some may try to distinguish between “statements of fact that are observed and written by a human and those that are collected mechanically,” but such a distinction would be “untenable.” Bambauer, Is Data Speech?, supra note 84, at 59–60. On the other hand, Enrique Gimenez argues that no constitutional right for surreptitious newsgathering exists, nor does he believe that such a right should exist. Enrique J. Gimenez, Comment, Who Watches the Watchdogs?: The Status of Newsgathering Torts Against the Media in Light of the Food Lion Reversal, 52 Ala. L. Rev. 675, 676 (2001).

As Professor Jane Bambauer notes, “To this day, the right to access information is underdeveloped.” 101 Bambauer, Is Data Speech?, supra note 84, at 86. The case law indicates that courts believe newsgathering or information gathering is conduct incidental enough to speech that some constitutional protection must be granted, yet they are hesitant to develop the doctrine further. Under the spectrum theory that scholars have posited, the attenuation of the link between the conduct and the speech act is key. 102 See supra notes 87–91 and accompanying text (describing the spectrum theory). However, even in cases in which the First Amendment is implicated, lower courts that have addressed the issue have focused on considerations of privacy and used tort law as an exter­nal limit to information gathering, though in practice this limit seems lenient.

2. False Speech and Misrepresentations. — The act of lying is another form of conduct that implicates speech. 103 See, e.g., Animal Legal Def. Fund v. Wasden, 878 F.3d 1184, 1190 (9th Cir. 2018) (treating lies without material gain as speech rather than conduct). The First Amendment protects some, but not all, types of lies. 104 See United States v. Alvarez, 567 U.S. 709, 722, 729–30 (2012) (plurality opinion) (detailing how there has been false speech that has historically gone without First Amendment protection, yet extending protection to cover lies about military honors); Alan K. Chen & Justin Marceau, High Value Lies, Ugly Truths, and the First Amendment, 68 Vand. L. Rev. 1435, 1437 (2015) [hereinafter Chen & Marceau, High Value Lies] (noting that some lies, such as perjury and fraud, do not receive constitutional protection though other lies, such as “intentionally lying about military honors,” do). The Supreme Court has rejected First Amendment challenges to fraud claims, 105 Illinois ex rel. Madigan v. Telemarketing Assocs., Inc., 538 U.S. 600, 605, 612 (2003) (holding that “the First Amendment does not shield fraud” in the case of a charity that was sued for making false and misleading statements to donors). criminalization of one’s misrep­resenting status as a government official, 106 United States v. Lepowitch, 318 U.S. 702, 703–04 (1943). and regulations about false or misleading commercial speech. 107 Cent. Hudson Gas & Elec. Corp. v. Pub. Serv. Comm’n, 447 U.S. 557, 563 (1980). Traditionally, the First Amendment had not protected false statements if those statements were valueless. 108 See Alvarez, 567 U.S. at 739 (Alito, J., dissenting) (noting that there is “a long line of cases recognizing that the right to free speech does not protect false factual statements that inflict real harm and serve no legitimate interest”). Yet the Court’s decision in United States v. Alvarez makes clear that false speech must be more than merely valueless to be restricted—it must be fraudulent, defamatory, or cause some other “legally cognizable harm.” 109 Id. at 719, 722 (plurality opinion); see also Chen & Marceau, High Value Lies, supra note 104, at 1452 (“Alvarez, then, reflects a turning point: an intentional lie of little or no value, which arguably caused some harm, was nonetheless deemed protected.”). With the move away from the traditional inquiry based on the speech’s value 110 Previously, the Supreme Court had said that lies are not protected when they have no value and can cause harmful, real effects. See Chaplinsky v. New Hampshire, 315 U.S. 568, 571–72 (1942). and the Court’s “fractured” opinion, the exact analysis to apply in false speech cases is uncertain. 111 See Chen & Marceau, High Value Lies, supra note 104, at 1452 (detailing the Alvarez opinion). The Alvarez Court did stress, however, a desire to avoid chilling speech with governmental censorship 112 Alvarez, 567 U.S. at 722–24 (explaining that giving the government power to censor untruthful speech would “cast[] a chill, a chill the First Amendment cannot permit if free speech, thought, and discourse are to remain a foundation of our freedom”); id. at 733 (Breyer, J., concurring in the judgment, joined by Kagan, J.) (“[T]he threat of criminal prosecution for making a false statement can inhibit the speaker from making true state­ments, thereby ‘chilling’ a kind of speech that lies at the First Amendment’s heart.”). —unless such speech “was used to gain a material advantage.” 113 Id. at 723 (plurality opinion).

The Supreme Court has yet to address “the relevant constitutional implications of a common law misrepresentation action directed against a media defendant,” 114 Veilleux v. Nat’l Broad. Co., 206 F.3d 92, 126 (1st Cir. 2000). so analyzing these questions requires recourse to lower court doctrine. When it comes to misrepresentations to gain access for an investigative purpose, the Ninth Circuit in Animal Legal Defense Fund v. Wasden relied on Alvarez to invalidate portions of Idaho’s “Ag-Gag law[],” which made it a crime to, among other things, use misrepresen­tations to knowingly enter an agricultural facility, obtain records, obtain employment with the intent to cause economic or other injury, or make recordings via misrepresentation. 115 878 F.3d 1184, 1189–91 (9th Cir. 2018). The court held that Idaho could not criminalize entry by misrepresentation, for entry onto property, even private property, is not a material gain. 116 Id. at 1194 –95. In justifying its reasoning, the court referred to Food Lion and Desnick, cases in which courts held that testers and undercover journalists had not committed any tort violations. 117 Id. at 1196; see also supra notes 96–99 for an overview of Desnick and the cases that followed it, including Food Lion. In fact, the Desnick court explicitly stated there was no fraud in a tester scheme because “a scheme to expose publicly any bad practices that the investiga­tive team discovered” was not fraudulent. 118 Desnick v. Am. Broad. Cos., 44 F.3d 1345, 1355 (7th Cir. 1995).

On the other hand, the Wasden court held that there was no First Amendment bar to Idaho’s criminalization of misrepresentations made to obtain records or misrepresentations made to obtain employment when the speaker intended to cause economic or other injury to the facility, because both records and employment are material gains. 119 Wasden, 878 F.3d at 1199–203. The Ninth Circuit stated that while it may be true that undercover journalists seek to expose public threats rather than secure material gain, because the Supreme Court had specifically referred to offers for employment when giving examples of material gains and because undercover investi­gators are paid by the facilities as part of their employment, they are not immune from Alvarez’s holding. 120 Id. at 1201– 02. The court stressed, however, that “this does not mean that every investigative reporter hired under false pretenses intends to harm the employer,” and therefore, so long as the requisite intent is not present, journalists would be protected. 121 Id.

Thus, though the Ninth Circuit only partially invalidated Idaho’s ag-gag law, it nonetheless expressed caution in allowing criminalization based on misrepresentations, evidenced by its strict reading of what constituted “material gains” and its emphasis on the intent requirement for criminal­ization of misrepresentations for employment. 122 Id. Given this, while the Supreme Court has not addressed the specific issue of misrepresentation claims and newsgathering or civil rights testing, lower courts before and after Alvarez seem wary of validating tort claims under a misrep­resentation theory.

After reviewing the issue of algorithm-based discrimination, civil rights testing historically and online, and relevant doctrinal strands of First Amend­­ment case law, this Note next turns to the conflict between the First Amendment and the CFAA.

II. Tension Between the CFAA’s Prohibition on Online Civil Rights Testing and the First Amendment

In late June of 2016, the ACLU filed a complaint on behalf of aca­demic researchers and First Look Media Works 123 Complaint, supra note 64. challenging the consti­tutionality of the CFAA. 124 18 U.S.C. § 1030 (2012). The plaintiff researchers in Sandvig v. Sessions are actively attempting to research whether housing and hiring websites are engaging in discriminatory conduct, 125 Complaint, supra note 64, at 5–6. and First Look Media Works is a journalism platform hoping to conduct similar research on websites’ discriminatory use of data and algorithms. 126 Id.

The complaint alleges some possible First Amendment grounds on which the CFAA could be invalidated as a result of its prohibition on online testing and raises both facial and as-applied challenges. 127 Id. at 43–46. Though the plaintiffs’ facial challenge was dismissed, 128 Sandvig v. Sessions, No. 16-1368, 2018 WL 1568881, at *21–22 (D.D.C. Mar. 30, 2018). the issue remains that insofar as the CFAA restricts these plaintiffs’ rights to test for online discrimination, this law runs up against the First Amendment because “research, testing, or investigations” constitute protected speech or expres­sion. 129 Complaint, supra note 64, at 33–38. In response, the government filed a motion to dismiss, arguing that the conduct in this case was nonspeech activity not reached by the First Amendment. 130 Memorandum of Points & Authorities in Support of Defendant’s Motion to Dismiss at 20, Sandvig v. Sessions, No. 1:16-cv-1368 (D.D.C. filed Sept. 9, 2016), 2016 WL 5436529.

Though scholars have commented on the possible friction between the CFAA and the First Amendment for some years now, the particular issue raised by the plaintiffs in Sandvig remains unanswered. Previous scholarship has largely revolved around the First Amendment implications of the CFAA’s proscription of “hacktivism,” an act that may be protected by the First Amendment. 131 The CFAA was originally passed to target hacking. The House Report on the CFAA explained that “the conduct prohibited is analogous to that of ‘breaking and entering’ rather than using a computer (similar to the use of a gun) in committing the offense,” and Congress was specifically motivated by the inadequacy of criminal laws in covering hacking. H.R. Rep. No. 98-894, at 20 (1984), as reprinted in 1984 U.S.C.C.A.N. 3686, 3706; see also Cyrus Y. Chung, The Computer Fraud and Abuse Act: How Computer Science Can Help with the Problem of Overbreadth, 24 Harv. J.L. & Tech. 233, 237–39 (2010) (detailing the legislative history behind the CFAA and the subsequent amendments act). However, recent scholarship has noted that hacking encompasses “hacktivism,” which is the act of hacking “to promote social or political ends or to effect social or political change.” Andrew T. Illig, Computer Age Protesting: Why Hacktivism Is a Viable Option for Modern Social Activists, 119 Penn St. L. Rev. 1033, 1036 (2015). These scholars also posit that because the First Amendment’s purposes are being served by hacktivism, it should be constitutionally protected activity. Id. at 1048–49. Other scholarship argues that the CFAA is an overbroad law that inadvertently infringes upon the protected speech of independent commentators online. 132 See, e.g., Philip F. DiSanto, Note, Blurred Lines of Identity Crimes: Intersection of the First Amendment and Federal Identity Fraud, 115 Colum. L. Rev. 941, 943 (2015). For instance, sharing a hyperlink to personal documents could be protected speech, yet this could open the sharer to criminal liability. Id. at 963–64. Additionally, the CFAA may be in tension with the First Amendment because it allows website owners to control information use and thus essentially confers private rights in information to the site owners, thereby restricting speech. 133 See Christine D. Galbraith, Access Denied: Improper Use of the Computer Fraud and Abuse Act to Control Information on Publicly Accessible Internet Websites, 63 Md. L. Rev. 320, 322 (2004). Finally, the most recent scholarship involving this topic has argued in theoretical terms that the First Amendment should encompass “Empirical Liberty,” or a “scientific” right to experiment and research. 134 See generally Jane R. Bambauer, The Empirical First Amendment, 78 Ohio St. L.J. 947 (2017). Unlike this Note, Professor Bambauer’s essay focuses on the theoretical underpinnings of the First Amendment and argues that the law should be committed to a scientific approach to First Amendment theory. Id. at 948–49.

Still, the scholarship exploring the tensions between the First Amend­ment and the CFAA is generally fairly limited in scope. It has not addressed the specific question raised by Sandvig v. Sessions: whether the CFAA infringes on the First Amendment rights of testers by preventing online civil rights testing. The remainder of this Note turns to this issue and contends that the CFAA conflicts with the First Amendment when applied to advocates, journalists, and researchers seeking to conduct online civil rights testing because such testing is protected by the First Amendment.

This Part first examines the CFAA and highlights how courts’ inter­pretation of this statute has resulted in the CFAA’s effective prohibi­tion against online civil rights testing. The following section then out­lines the various doctrinal bases that litigators could use to seek First Amendment protection for online civil rights testing and assesses the via­bility of each one, ultimately arguing that the theories of false speech and conduct incidental to speech can give rise to a First Amendment right for testers. Therefore, insofar as the CFAA is applied against online civil rights testers, such application infringes upon the First Amendment.

A. The Computer Fraud and Abuse Act (CFAA)

Congress passed the Computer Fraud and Abuse Act 135 18 U.S.C. § 1030 (2012). in 1984 to address the issue of cybersecurity. 136 S. Rep. No. 104-357, at 3–4 (1996) (noting that the CFAA was passed “to provide a clear statement of proscribed activity concerning computers” and that it “generally pro­hibits the unauthorized use of computers to obtain classified or private financial record information, to trespass in Federal Government computers, to commit frauds, or to transmit harmful computer viruses”); see also Orin S. Kerr, Norms of Computer Trespass, 116 Colum. L. Rev. 1143, 1144 (2016). The CFAA prohibits computer trespass by preventing parties from knowingly accessing a protected computer 137 What constitutes a protected computer is “effectively all computers with Internet access.” United States v. Nosal, 676 F.3d 854, 859 (9th Cir. 2012) (en banc). either “without authorization” or by “exceed[ing] authorized access.” 138 18 U.S.C. § 1030(a)(4). The scope of the CFAA is still an open question, with courts “divided deeply over when access is authorized.” 139 Kerr, supra note 136, at 1144–45; see also Note, The Vagaries of Vagueness: Rethinking the CFAA as a Problem of Private Nondelegation, 127 Harv. L. Rev. 751, 751–52 (2013). Defendants have been prosecuted under the CFAA for acts ranging from gathering information from an employer’s computer for personal use in violation of a company rule 140 United States v. Rodriguez, 628 F.3d 1258, 1260 (11th Cir. 2010). to collecting information from a website after the site had sent a cease-and-desist letter and blocked the defendant’s IP address. 141 Craigslist Inc. v. 3Taps Inc., 942 F. Supp. 2d 962, 968–70 (N.D. Cal. 2013).

Some courts have also held that violation of a website’s terms of ser­vice can trigger the CFAA, 142 See, e.g., CollegeSource, Inc. v. AcademyOne, Inc., 597 F. App’x 116, 129–30 (3d Cir. 2015); EF Cultural Travel BV v. Zefer Corp., 318 F.3d 58, 62 (1st Cir. 2003); United States v. Lowson, No. 10-114, 2010 WL 9552416, at *1 (D.N.J. Oct. 12, 2010); United States v. Drew, 259 F.R.D. 449, 461–62 (C.D. Cal. 2009); Sw. Airlines Co. v. Farechase, Inc., 318 F. Supp. 2d 435, 439–40 (N.D. Tex. 2004); Am. Online, Inc. v. Nat’l Health Care Discount, Inc., 174 F. Supp. 2d 890, 899 (N.D. Iowa 2001); Register.com, Inc. v. Verio, Inc., 126 F. Supp. 2d 238, 247–51 (S.D.N.Y. 2000). and the Department of Justice has supported this position. 143 See Office of Legal Educ., Dep’t of Justice, Prosecuting Computer Crimes 8–9 (2015), http://www.justice.gov/sites/default/files/criminal-ccips/legacy/2015/01/14/ccmanual.pdf [http://perma.cc/D8G4-66GH] (“It is relatively easy to prove that a defendant had only limited authority to access a computer in cases where the defendant’s access was limited by restrictions that were memorialized in writing, such as terms of service . . . .”). The rationale is that website terms of service are meant to limit users’ access, so users who violate those terms thereby exceed “authorized” access. 144 See, e.g., EF Cultural Travel BV, 318 F.3d at 62. Thus, this reading gives effect to website terms of service that prohibit the “sock puppet” or “scraping” online civil rights testing methods. For instance, websites, including those that provide housing and employment services, such as Zillow.com 145 Zillow, supra note 16 (“Automated queries (including screen and database scraping, spiders, robots, crawlers and any other automated activity with the purpose of obtaining information from the Services) are strictly prohibited on the Services . . . .”). and Indeed.com, 146 Cookies, Privacy and Terms of Service, Indeed, http://www.indeed.com/legal [http://perma.cc/87ZY-SRSC] (last updated Sept. 15, 2016) (noting that users cannot, among other activities, “create user accounts by automated means or under false or fraud­ulent pretenses”). prohibit the use of bots, scraping, and the creation of false accounts. 147 Complaint, supra note 64, at 18–19. Many companies’ terms of service include provisions that would prohibit testers from using certain types of testing; for some examples, see supra notes 145–146. These sites could be engaging in algorithm-based discrimination, yet absent an exception for civil rights testing, testers using either the sock puppet or scraping method to detect such discrimination violate the sites’ terms of service and could thus face liability under the CFAA.

To protect online testing, the plaintiffs in Sandvig v. Sessions seek to strike down the CFAA on First Amendment grounds. 148 See Complaint, supra note 64, at 46. Note that Plaintiffs sought to invalidate on multiple grounds, one of which is the First Amendment. See generally id. This Note next argues that because the First Amendment can in fact protect online civil rights testing, the CFAA, under courts’ current interpreta­tion, infringes on constitutionally protected activities.

B. The Incompatibility Between the First Amendment and a CFAA Testing Restriction

Whether the CFAA conflicts with the First Amendment because of its restriction on online civil rights testing is predicated on a few issues, 149 See supra note 21 (noting the issue of whether the state action requirement for First Amendment scrutiny has been met). the most important of which is whether civil rights testing of this kind is even conduct that could be protected by the First Amendment. 150 See City of Dallas v. Stanglin, 490 U.S. 19, 25 (1989) (“It is possible to find some kernel of expression in almost every activity a person undertakes—for example, walking down the street or meeting one’s friends at a shopping mall—but such a kernel is not sufficient to bring the activity within the protection of the First Amendment.”); see also Kreimer, supra note 73, at 370 (“[I]t is common currency that not all actions can claim First Amendment protection.”); Post, Recuperating, supra note 73, at 1250 (“Like any legal provision, the [First] Amendment must contain threshold conditions that specify when its particular doctrines and values will be activated and applied.”). This section will look at the various doctrinal lines that could offer protection for testing conduct and assess the strength of each one. Section II.B.1 evaluates whether the First Amendment’s protection for false speech con­flicts with the testing prohibition at issue here, and section II.B.2 grounds a claim for First Amendment protection in the idea that civil rights testing can be seen as conduct incidental to speech.

1. False Speech. — First is the argument that the CFAA, when applied to online civil rights testing, violates the First Amendment because it im­poses criminal penalties on misrepresentation, or false speech, which is protected by the First Amendment in certain situations. Though online testing is technically conduct, it nonetheless can be seen as an act of false speech. Just as in-person testers make affirmative misrepresentations about their identities or intentions, online testers do essentially the same: The testers (or the bots they program) paint themselves as individuals who they are not for the purpose of gaining access to certain sites. 151 The primary difference is that in-person testers might orally misrepresent themselves, while online testers do so virtually. Yet it’s unclear why this difference should be material—testers make misrepresentations in either case, and to say that spoken misrepresentations are protected while “typed out” or other nonverbal misrepresentations are not seems irrational. Though this calculus might change if the bots are the ones misrepresenting themselves, it’s not clear whether such an argument would be supported given the current case law. See infra note 239 and accompanying text (discussing whether the fact that bots are en­gaging in the speech act in question defeats a First Amendment claim). Due to these parallels, it is sensible to apply the false speech doctrine when ana­lyzing claims of testers under the First Amendment.

Alvarez held that the First Amendment prevents criminalization of falsehoods that are not fraudulent or cause no tangible harm, 152 See supra notes 109–113 and accompanying text (explaining the impact of the Alvarez decision). and the CFAA’s prohibition on testing conflicts with this holding. 153 This argument relies on the validity of Desnick v. Am. Broad. Cos., 44 F.3d 1345, 1355 (7th Cir. 1995) (stating that a tester scheme aimed to expose bad practices discovered by investigators was “not a fraudulent scheme”). Testing activity can be seen as an act of false speech by way of making a misrepresenta­tion, as testers misrepresent their identities and intentions. Yet courts have found that the kind of misrepresentation traditional testers engage in is not fraudulent or harmful. 154 Apple Corps Ltd. v. Int’l Collectors Soc’y, 15 F. Supp. 2d 456, 476 (D.N.J. 1998) (finding that the misrepresentation of lawyers who were acting as testers did not manifest wrongdoing such as fraud). Furthermore, recent applications of Alvarez at the lower court level in ag-gag cases, such as Animal Legal Defense Fund v. Wasden, 155 878 F.3d 1184, 1194 (9th Cir. 2018) (noting that portions of Idaho’s ag-gag law violate the First Amendment by criminalizing misrepresentations made for entry and under­cover recording). also lend support to the Sandvig plaintiffs. Just as these ag-gag laws criminalize misrepresenting oneself as a worker or potential employee for the purposes of undercover investigations, 156 Id. at 1190. the CFAA effec­tively criminalizes misrep­resenting oneself as a potential customer, tenant, or job applicant even for the purposes of online civil rights testing.

In Wasden, the Ninth Circuit said that misrepresentations made for entry were not made for material gain. 157 Id. at 1195–98. Though the court said that misrepresentations made for employment were made for material gain, the court emphasized that the statute in this case only criminalizes mis­representations made with the intent to cause harm to the facility and therefore was not overbroad. 158 Id. at 1201– 03.

Protecting the misrepresentations made in online civil rights testing does not run afoul of the Ninth Circuit’s decision. Though it could be argued that employment discrimination testers may seek to cause the discriminatory employers they test legal or economic harm, when testers initially seek employment opportunities, they typically are unsure of whether a particular employer is actually discriminatory. 159 Cf. Fred Freiberg, A Test of Our Fairness, 41 Urb. Law. 239, 241– 42 (2009) (explaining that testing is the first act that takes place after housing discrimination is reported and advocating for testing that is proactive and not just reactive). Thus, like an investiga­tive journalist, a tester in most cases will not apply for employ­ment with the preconceived intent to cause harm, and therefore, testers would not be subject to the Idaho statute as the Ninth Circuit described it. The case law accordingly supports the inference that the First Amend­ment protects the kind of investigative activity the plaintiffs in Sandvig seek to undertake.

Even if the misrepresentation for employment purposes involved in ag-gag law cases is not wholly clear-cut based on Alvarez, misrepresenta­tions made during civil rights testing should not be considered analogous to the type of misrepresentation for gain the Court has said would fall outside the protections of the First Amendment, because misrepresentations made by testers are not fraudulent. 160 See Va. State Bd. of Pharmacy v. Va. Citizens Consumer Council, Inc., 425 U.S. 748, 771 (1976) (holding that fraudulent statements are not protected by the First Amendment). Courts have typically held that promises made in the course of investigative reporting are not fraudulent. See, e.g., Food Lion, Inc. v. Capital Cities/ABC, Inc., 194 F.3d 505, 514 (4th Cir. 1999) (finding that, in a case in which undercover reporters were used by ABC, Food Lion could not maintain fraud because there was no injurious reliance); see also David B. Isbell & Lucantonio N. Salvi, Ethical Responsibility of Lawyers for Deception by Undercover Investigators and Discrimination Testers: An Analysis of the Provisions Prohibiting Misrepresentation Under the Model Rules of Professional Conduct, 8 Geo. J. Legal Ethics 791, 812 (1995) (explaining that the “misrepresentations of the mild sort necessarily made by discrimination testers and undercover investigators” are not akin to conduct involving dishonesty, fraud, deceit, or tortious misrepresentation). A rule that misrepresentations made during civil rights testing are actionable as fraud would have a chilling effect on all kinds of testing and undermine the important pur­pose of testers, which has been repeatedly recognized by the courts. 161 See supra section I.A.2 (discussing the history of civil rights testers). To that end, some district courts have explicitly recognized that testers posing as potential consumers or applicants do not engage in misrepresentation. 162 See Apple Corps Ltd. v. Int’l Collectors Soc’y, 15 F. Supp. 2d 456, 476 (D.N.J. 1998).

The most significant drawback to grounding testers’ First Amendment rights in the false speech framework is the ambiguity presented by Animal Legal Defense Fund v. Wasden. Though the court seemed to suggest that any misrepresentations made for employment are unprotected, it also repeatedly emphasized the specific intent requirement built into the statute and rebutted the Animal Legal Defense Fund’s argument that the statute was overbroad by pointing to the intent requirement. 163 Wasden, 878 F.3d at 1201–02. Given the court’s treatment, it remains somewhat unclear whether the First Amend­ment would allow a categorical ban on misrepresentations for employ­ment purposes, regard­less of the speaker’s intentions. 164 In another ag-gag case at the district court level, the court noted that an argu­ment based on Alvarez upholding the criminalization of obtaining an offer of employment through false speech “might carry some weight.” Animal Legal Def. Fund v. Herbert, 263 F. Supp. 3d 1193, 1206 (D. Utah 2017). But again, the court there was ambiguous and struck down the provision at issue because it unlimitedly prohibited false speech used to gain access to agricultural facilities. Id. Were courts to allow prohibitions on false speech for employment purposes irrespective of whether the speaker had a truth-seeking intent, the result would cause some internal tension within First Amend­ment law: In Alvarez, the defend­ant engaged in speech for no valuable purpose and yet his speech merited protection, 165 United States v. Alvarez, 567 U.S. 709, 714 (2012) (“For all the record shows, respondent’s statements were but a pathetic attempt to gain respect that eluded him.”). whereas those who engage in misrepresentations to examine potential legal violations are attempting to discover the truth and thereby per­form a public service but would not be protected. Given the First Amend­­ment’s strong roots in protecting truth-seeking and encouraging public knowledge and debate, 166 See Chen & Marceau, High Value Lies, supra note 104, at 1437–38; see also James L. Oakes, Tolerance Theory and the First Amendment, 85 Mich. L. Rev. 1135, 1137 (1987) (reviewing Lee C. Bollinger, The Tolerant Society: Free Speech and Extremist Speech in America (1986)) (arguing that a classical model of First Amendment theory finds that its “values include truth-seeking and knowledge-advancement”). allowing for unqualified prohibitions on misrep­resentations made for employment seems antithetical to the amendment.

Even if blanket bans on misrepresentations made for offers of em­ployment were to be upheld, online civil rights testers may still be exempt. Unlike undercover investigators who seek to research an entity or industry by working within it, testers merely aim to assess whether differential treatment occurs in the hiring process. 167 Michael J. Yelnosky, Filling an Enforcement Void: Using Testers to Uncover and Remedy Discrimination in Hiring for Lower-Skilled, Entry-Level Jobs, 26 U. Mich. J.L. Reform 403, 437 (1993) (noting that employment testers have no interest in the actual employment opportunity they apply for). Therefore, testers aren’t really seeking offers of employment. In addition, the Wasden court focused on the fact that “undercover investigators are nonetheless paid by the agricultural production facility as part of their employ­ment.” 168 Wasden, 878 F.3d at 1202. But civil rights testers would not actually be paid, as they are not planning on working for the prospective employer. Thus, were a categorical prohibition on misrepresentations made for employment to be upheld, it’s nonetheless feasible to protect testers.

Further counterarguments against protecting civil rights testers based on false speech doctrine can also be adequately addressed. First, some may argue that the Stolen Valor Act of 2005, which was at issue in Alvarez, directly criminalizes false speech just as ag-gag laws do, 169 Alvarez, 567 U.S. at 713 (describing the Stolen Valor Act). whereas the CFAA imposes no such facial proscription. However, as applied in the case of civil rights testers, the effect is the same: The courts’ current interpre­tation of the CFAA effectively prevents testers from engaging in false speech even if it does not result in fraud or material gain. Thus, though the CFAA is not a facial proscription on speech, it nonetheless runs afoul of the First Amendment when it is interpreted to block tester activity.

For these reasons, under the current framework of the false speech doctrine, the First Amendment likely could protect misrepresentations made by testers in the course of their conduct. This, in turn, creates tension between the CFAA and freedom of speech—an interpretation of the CFAA that renders online testing illegal and thereby penalizes false speech runs counter to Alvarez’s holding.

2. Conduct Incidental to Speech. — The second line of doctrine that could offer an avenue for First Amendment safeguarding is that of con­duct incidental to speech. Within this case law, there are two different approaches that could be used. First, there is the right of access line of cases that follows Richmond Newspapers, Inc. v. Virginia. 170 448 U.S. 555 (1980); see also supra notes 78–83 and accompanying text (discussing Richmond Newspapers and its progeny). Second, there is the more general doctrine of conduct incidental to speech. 171 See supra notes 84–86 and accompanying text (discussing a general right to newsgather). Although there is no formula for analyzing what constitutes conduct incidental to speech, perhaps because the Court has been hesitant to flesh out this right, 172 See Houchins v. KQED, 438 U.S. 1, 8–12 (1978) (reading precedent on a right to newsgather strictly). this section will use the spectrum analysis discussed by Professors Chen and Marceau, 173 See supra notes 87–91 (explaining that conduct incidental to speech falls on a spectrum). along with lower court decisions, to assess the claim that testing conduct could be seen as incidental enough to speech to garner First Amendment protection. It also discusses potential barriers and the strengths of the analysis under each approach. Though it seems unlikely that Richmond Newspapers and its progeny could support a First Amendment right to conduct civil rights testing, the right can be firmly grounded in the doctrine protecting conduct incidental to speech.

a. First Amendment Right of Access. — While the argument for tester protection based on the line of right of access cases following Richmond Newspapers is plausible, it is unlikely to succeed. Such an argument would be grounded in the two considerations set forth by the Supreme Court: history and the role of public access. 174 Press-Enter. Co. v. Superior Court, 478 U.S. 1, 8 (1986) (formulating a two-part test for right of access, first looking at “whether the place and process have historically been open to the press” and then looking at “whether public access plays a significant positive role in the functioning of the particular process in question”). Historically, civil rights testing occurred in business operations that not only were open to the general public but actually relied on the public coming in and out. 175 Usually, in-person testers are used to assess whether discrimination occurs in places such as housing operations, see Freiberg, supra note 159, at 240, and other businesses, see Haydon, supra note 56, at 1216 (describing a typical test as occurring in public accom­modations like restaurants), all of which seem to need public patronage. These busi­nesses are not public areas in the sense that they are publicly or govern­mentally owned, but they have been historically open to public access. 176 See McMahon v. City of Panama City Beach, 180 F. Supp. 3d 1076, 1096 (N.D. Fla. 2016) (finding that an event in a park that occurred pursuant to an exclusive permit held by a for-profit entity but that did not have any “no trespassing” signs was a public forum). But see Sw. Cmty. Res., Inc. v. Simon Prop. Grp., 108 F. Supp. 2d 1239, 1257 (D.N.M. 2000) (finding that the defendant’s shopping mall did not dedicate itself to public use because there was no invitation to use it for expressive activity). It is also worth noting that even though civil rights testers are not necessarily members of the press, they would still be able to claim a right of access under the First Amendment. See Richmond Newspapers, Inc. v. Virginia, 448 U.S. 555, 580 (1980) (concluding that the public has the right to access the trial in question under the First Amendment). With regard to the second prong, it could be argued that civil rights testing plays a significant role in the functioning of businesses and com­merce. 177 See Heart of Atlanta Motel, Inc. v. United States, 379 U.S. 241, 250 (1964) (noting that the Civil Rights Act of 1964, which includes public accommodations provisions, was passed pursuant to the Commerce Clause). In fact, passage of Title II of the Civil Rights Act, which prohibits dis­crimination in public accommodations, evidences the economic benefits of antidiscrimination, though the statute can be justified on both moral and economic grounds. 178 For an overview of the economic justifications for Title II of the Civil Rights Act, see Harry T. Quick, Public Accommodations: A Justification of Title II of the Civil Rights Act of 1964, 16 Case W. Res. L. Rev. 660, 664–65 (1965).

Yet a court would be unlikely to accept this argument because courts have not yet applied this test to extragovernmental contexts. 179 See supra note 83 and accompanying text; see also Paul Coppock, Doors to Remain Open During Business Hours: Maintaining the Media’s (and Public’s) First Amendment Right of Access in the Face of Changing Technology, 58 S.D. L. Rev. 319, 320 (2013) (“United States Supreme Court cases dealing with the issue of public access have been in the con­text of criminal proceedings.”). Richmond Newspapers focused on the historical, rather than functional, reasons for allowing a right of access, and the history largely revolved around the First Amendment’s “common core purpose of assuring freedom of communication on matters relating to the functioning of government.” 180 Richmond Newspapers, Inc., 448 U.S. at 556, 572, 575 (emphasis added). If the First Amendment purpose being served is the ability to use speech as a check on government actions, limiting the progeny of Richmond Newspapers to governmental affairs makes sense.

The Supreme Court recently decided a case involving access and the First Amendment—but rather than dealing with access to a proceeding or physical place, the case dealt with a sex offender’s right to access the internet. 181 Packingham v. North Carolina, 137 S. Ct. 1730, 1733 (2017). Though the case primarily addressed the interplay between the internet and the public forum doctrine, 182 For a discussion on this aspect of the case, see infra notes 233–237. the Court none­theless seemed to ground the right to access in the speech acts that the access facilitated. 183 See Packingham, 137 S. Ct. at 1737 (“In sum, to foreclose access to social media altogether is to prevent the user from engaging in the legitimate exercise of First Amend­ment rights.”). The Court’s decision thus suggests that the doctrinal line in­volving conduct incidental to speech is a stronger basis for pro­tecting civil rights testing, as discussed next.

b. Spectrum Analysis. — A more viable claim is that civil rights testing is protected as a kind of newsgathering or information-gathering conduct that falls close enough to speech to merit protection, even if there is no “unrestrained right” to gather information. 184 Zemel v. Rusk, 381 U.S. 1, 17 (1965). Due to the lack of guidance from the Supreme Court on when newsgathering actions merit protection, 185 See S.H.A.R.K. v. Metro Parks Serving Summit Cty., 499 F.3d 553, 560 (6th Cir. 2007) (noting that the Supreme Court has not established a “clearly defined framework” for right-of-access-to-information cases outside of the judicial proceedings context). this discussion will instead use lower court opinions and the spectrum theory posited by Professors Chen and Marceau. 186 See supra notes 87–91 and accompanying text (introducing the spectrum theory).

Under the spectrum theory, conduct that falls closer to a final speech act is more likely to be protected. 187 Marceau & Chen, Free Speech, supra note 87, at 1018–19. With regard to civil rights testing, the level of attenuation between the speech activity (the publica­tion of the testing results) and the conduct seeking protection (the testing itself) is insubstantial. In the context of an exposé, what is more directly antecedent to its writing and subsequent publication than the investiga­tion that yielded its content? If campaign financing is seen as necessary enough to political speech to merit First Amendment protection, it makes little sense to say that civil rights testing is unnecessary for an investigative news or research article to be published, since it is this very testing that provides the content of such an article. Further, as Professor Bambauer notes, a rule that data are speech while their creation is not, when applied to other contexts, “would immediately draw suspicion: Selling books is protected speech; could printing them possibly be regulable conduct? Music is speech, but could it be that recording and producing an album is conduct?” 188 Bambauer, Is Data Speech?, supra note 84, at 79–80. Professor Bambauer goes on further to argue that a ban on photography should attract First Amendment scrutiny because the “very purpose of a photography ban is to prevent a wider audience from seeing the scene.” Id. at 83.

This theory has most recently reemerged in the form of “right to record” cases at the appellate level. Every circuit that has considered whether the First Amendment protects the right of individuals to record official police activity has answered in the affirmative. 189 To date, the First, Third, Fifth, Ninth, and Eleventh Circuits have held that citizens have the right to record police activity, while the Seventh Circuit has held that there is a right to intercept communications of public officials engaged in their public duties. Fields v. City of Philadelphia, 862 F.3d 353, 356 (3d Cir. 2017); Turner v. Lieutenant Driver, 848 F.3d 678, 690 (5th Cir. 2017); Gericke v. Begin, 753 F.3d 1, 7 (1st Cir. 2014); ACLU v. Alvarez, 679 F.3d 583, 597–600 (7th Cir. 2012); Smith v. City of Cumming, 212 F.3d 1332, 1333 (11th Cir. 2000); Fordyce v. City of Seattle, 55 F.3d 436, 439 (9th Cir. 1995). These cases have rejected the artificial distinction between the videos or recordings them­selves, which are unquestionably protected by the First Amendment, 190 See Brown v. Entm’t Merchs. Ass’n, 564 U.S. 786, 790 (2011). and the act of making the recordings or videos. As the Third Circuit said, in order to give actual photos, videos, and recordings meaningful protec­tion, “the Amendment must also protect the act of creating that material.” 191 Fields, 862 F.3d at 358. The court went on to explicitly recognize that there is no practical dis­tinction between preventing citizens from making a recording and banning the possession or distribution of recordings. 192 Id. Other “right to record” cases have echoed this understanding that both the final speech act and its creation should be equally protected. 193 See, e.g., Animal Legal Def. Fund v. Wasden, 878 F.3d 1184, 1203 (9th Cir. 2018) (“It defies common sense to disaggregate the creation of the video from the video or audio recording itself.”); Turner, 848 F.3d at 688–89 (noting that the corollary to the principle that the First Amendment protects film is that “the First Amendment protects the act of making film”); Glik v. Cunniffe, 655 F.3d 78, 82 (1st Cir. 2011) (recognizing that “the First Amendment’s aegis extends further than the text’s proscription on laws ‘abridging the freedom of speech, or of the press,’ and encompasses a range of conduct related to the gathering and dissemination of information” in protecting a right to record). More recently, lower courts in circuits that have not weighed in on this question have nonetheless recog­nized First Amendment protections for recording under the theory that such action is a necessary step toward expressive activity. See, e.g., Animal Legal Def. Fund v. Herbert, 263 F. Supp. 3d 1193, 1208 (D. Utah 2017) (“Because recordings themselves are protected by the First Amendment, so too must the making of those recordings be protected.”); Higginbotham v. City of New York, 105 F. Supp. 3d 369, 378 (S.D.N.Y. 2015) (“While videotaping an event is not itself expressive activity, it is an essential step towards an expressive activity, at least when performed by a professional journalist who intends, at the time of recording, to disseminate the product of his work.”); Crawford v. Geiger, 996 F. Supp. 2d 603, 614 (N.D. Ohio 2014) (noting that the First Amendment protects newsgathering in recognizing a right to “openly film police officers carrying out their duties”). But see Maple Heights News v. Lansky, No. 1:15CV53, 2017 WL 951426, at *3 (N.D. Ohio Mar. 10, 2017) (limiting Crawford to the right to record police officers performing their duties). Thus, courts have seen pro­hibitions on recording as “necessarily limit[ing] the information that might later be published or broadcast—whether to the general public or to a single family member or friend—and thus burden[ing] First Amendment rights.” 194 ACLU v. Alvarez, 679 F.3d at 597–600. In ACLU v. Alvarez, the Seventh Circuit reversed the denial of a preliminary injunction against a state wiretapping law. Id. Professor Bambauer argues that all-party-consent wiretap statutes, like the one at issue in ACLU v. Alvarez, “have already begun to fall apart under the weight of increased judicial scrutiny.” Bambauer, Is Data Speech?, supra note 84, at 83.

Another weighty consideration pushing these courts to recognize a “right to record” was the notion that such recordings may assist in unveil­ing and correcting police misconduct. As one court noted, civilian videos have filled the gaps when recordings of police conduct are otherwise unavailable or withheld, and these videos have been critical in kick-starting reform and assisting civil rights investigations. 195 Fields, 862 F.3d at 359–60; see also Turner, 848 F.3d at 689 (“Filming the police contributes to the public’s ability to hold the police accountable, ensure that police officers are not abusing their power, and make informed decisions about police policy.”); Glik, 655 F.3d at 82–83 (highlighting the uncovering of abuses and improvement of government functions as benefits of recognizing a right to record). Even absent publication of the videos, the act of recording can improve policing. 196 Fields, 862 F.3d at 359–60.

Under a similar logic, lower courts have gone one step further and recognized that investigations are incidental to speech. One court found that a plaintiff’s evidence-gathering activities, which involved monitoring and photographing the city’s mayor, were protected conduct because they gave rise to her speech concerning the mayor and police department’s actions. 197 Pomykacz v. Borough of West Wildwood, 438 F. Supp. 2d 504, 512–13 (D.N.J. 2006). Similarly, a California appellate court found that hiring an investigator could be protected as conduct that “intrinsically facilitates exercise of free speech.” 198 Tichnin v. City of Morgan Hill, 99 Cal. Rptr. 3d 661, 681 (Ct. App. 2009). Even before this case, a California appellate court had found that the creation of “surreptitious recordings which were used in connection with an investigative report by the media” was an act in fur­ther­ance of First Amendment free speech rights. Lieberman v. KCOP Television, Inc., 1 Cal. Rptr. 3d 536, 541 (Ct. App. 2003). Another court extended First Amend­ment pro­tection to the conduct of interviewing a prisoner as part of the plaintiff’s project of writing the biography of James Earl Ray, holding a Bureau of Prisons ban on interviews to be unconstitutional. 199 McMillan v. Carlson, 369 F. Supp. 1182, 1186, 1189 (D. Mass. 1973). Further, a court held that a statute prohibiting physicians from advertising their costs or fees, indirectly or directly, abridged the plaintiffs’ right to gather, receive, and publish such information. 200 Health Sys. Agency of N. Va. v. Va. State Bd. of Med., 424 F. Supp. 267, 275 (E.D. Va. 1976). Like in the “right to record” cases, these cases that more specifically deal with newsgathering similarly focus on the facilitative role of the investigative activities in question. These courts em­pha­size the “light [that] may be shed on” the information or ideas that will ultimately be published or broadcast 201 McMillan, 369 F. Supp. at 1186. and the “in further­ance of speech” element inherent in newsgathering. 202 Pomykacz, 438 F. Supp. 2d at 512–13; Lieberman, 1 Cal. Rptr. 3d at 541; see also Tichnin, 99 Cal. Rptr. 3d at 684 (noting that “the right of free speech protects not only the actual expressions of one’s views . . . but also non-expressive conduct that intrinsically facil­itates one’s ability to exercise the right of free speech”). These courts have also been careful to take note of the public interest in the final speech act. 203 ACLU v. Alvarez, 679 F.3d 583, 597 (7th Cir. 2012) (discussing how the statute at issue would prevent nonconsensual recording of a public official performing official duties in public, a topic of great public interest); Pomykacz, 438 F. Supp. 2d at 513 (noting the importance of a plaintiff’s ultimate speech act, which was political speech regarding the conduct of the mayor and police department); McMillan, 369 F. Supp. at 1186 (invoking the public’s interest in receiving information and ideas); Tichnin, 99 Cal. Rptr. 3d at 684 (detailing the information that plaintiff sought to investigate, “which, if true, would be a matter of public concern that [plaintiff] intended to communicate”); Lieberman, 1 Cal. Rptr. 3d at 541 (explaining that the topic of the broadcast at issue “is an issue of great public interest”).

These elements on which the “right to record” and newsgathering cases turn can also be found in the context of online civil rights testing. As highlighted above, testing activity, though conduct, is necessary for the documentation of the results, which is the final speech act; just as recording video or audio is a means to the final film or tape, online civil rights testing is the means by which the final speech act is produced. Regardless of whether the results are published, the means by which these results are procured should therefore receive First Amend­ment protec­tion. 204 See ACLU v. Alvarez, 679 F.3d at 597–600 (noting that the broadcast of a recording need not be publicized for the act of recording to be protected conduct). If prohibiting the publication of civil rights testing results would infringe on the First Amendment, then so too does a pro­hibition on the testing itself. To hold otherwise would create an end run around the Constitution.

Additionally, online civil rights testing generates significant public value. Investigations in general have played a crucial role in exposing the truth in certain industries and practices, dating as far back as the early 1900s with Upton Sinclair’s The Jungle 205 See generally Upton Sinclair, The Jungle (1906) (highlighting issues in the food industry through undercover investigation). See also Chen & Marceau, High Value Lies, supra note 104, at 1457 (“Sinclair’s work was critical to revealing the unsavory practices of a wealthy and powerful industry to public scrutiny.”). and the work of other muck­rakers. 206 Muckrakers were investigative journalists whose work highlighted social and political issues that needed reform, such as those of poverty, poor workers’ conditions, corporate misconduct, and political corruption. Laurie C. Hillstrom, Defining Moments: Muckrakers and the Progressive Era 3 (2009). Muckrakers are arguably one of the most heralded groups of journalists in history for their important work. See id. (“By bringing such issues to light, these watchdog journalists helped protect the American people—and U.S. democracy—from abuse at the hands of powerful interests.”). Civil rights testing more specifically, however, serves a public function in the same way that recording the police can: Whereas police recordings can uncover police misconduct, online civil rights testing can uncover violations of civil rights statutes. 207 See supra section I.A.2 (discussing the history and use of civil rights testers). As it currently stands, online civil rights testing methods such as sock puppet and scraping audits are the most viable methods of checking whether algorithms comply with antidis­crimination laws. 208 See Sandvig et al., supra note 5, at 12–14. Consequently, prohibiting this testing se­verely under­mines the ability to monitor companies transacting online. And without the safe harbor from testing that the CFAA affords these companies, they may become more conscious of the disparate effects of their technology and ameliorate these effects voluntarily, just as the act of recording police officers can itself mitigate against misconduct. 209 See Fields v. City of Philadelphia, 862 F.3d 353, 359–60 (3d Cir. 2017). More­over, publicizing the issues surrounding algorithms and discrimination can also lead to public discourse on matters such as the evolving stand­ards of civil rights enforcement or the drawbacks of current technology more broadly. 210 See W. Watersheds Project v. Michael, 869 F.3d 1189, 1197 (10th Cir. 2017) (emphasizing that plaintiffs plan to use their speech-creating activities to “further public debate” in holding that these activities are protected by the First Amendment). Thus, recognizing that online civil rights testing is protected conduct inci­dental to speech is squarely in line with the analysis of courts up­holding a “right to record” or a right to newsgather.

There are some features of these “right to record” and newsgathering cases that may make them appear distinguishable from the context of online civil rights testing. Yet more recent cases suggest that these distinc­tions are immaterial. For instance, one could argue that recording audio or video in preparation for broadcast is more analogous to writing in preparation for publishing; in other words, data collection seems to be more degrees removed than recording. Though perhaps technically true, courts have nonetheless begun treating data collection as protected con­duct, 211 Id. at 1192, 1195–96 (holding that statutes penalizing those who collect resource data target the creation of speech). Some courts recognized a right to gather data much earlier. For an example, see Health Sys. Agency of N. Va. v. Va. State Bd. of Med., 424 F. Supp. 267, 275 (E.D. Va. 1976) (striking down a prohibition on advertising physician costs because the prohibition would infringe on the plaintiff’s right to “gather, publish, and receive information”). suggesting that the difference between data collection and record­ing is inconsequential.

Second, these cases often involve public officials and public areas, 212 See, e.g., Western Watersheds Project, 869 F.3d at 1194 (assessing a statute that bur­dens data collection on public land); Fields, 862 F.3d at 359 (“The First Amendment protects the public’s right of access to information about their officials’ public activities.”); Glik v. Cunniffe, 655 F.3d 78, 82 (1st Cir. 2011) (“The filming of government officials engaged in their duties in a public place, including police officers performing their responsibilities, fits comfortably within [First Amendment] principles.”); Smith v. City of Cumming, 212 F.3d 1332, 1333 (11th Cir. 2000) (“The First Amendment protects the right to gather information about what public officials do on public property, and specifically, a right to record matters of public interest.”). while online civil rights testing takes place in a private domain. But courts reviewing recently enacted ag-gag laws nevertheless have struck down prohibitions on recording on private agricultural facilities, 213 Animal Legal Def. Fund v. Wasden, 878 F.3d 1184, 1203 (9th Cir. 2018) (holding that Idaho’s criminalization of recording agricultural production facilities without the consent of the owner violated the First Amendment); Animal Legal Def. Fund v. Herbert, 263 F. Supp. 3d 1193, 1208 (D. Utah 2017) (rejecting the State’s argument that there is no First Amendment right “because the Act applies only to speech on private property”). indicating that this too is an insignificant distinction. As one of these courts pointed out, the idea that speech acts on private facilities enjoy no First Amendment protection finds “no support in the case law.” 214 Herbert, 263 F. Supp. 3d at 1208. Rather, such an argument “confuses two related but distinct concepts: a land­owner’s ability to exclude from her property someone who wishes to speak, and the government’s ability to jail the person for that speech.” 215 Id. Thus, Utah’s criminalization of speech acts even on private property violated the First Amendment. 216 Id. The First Amendment issues in the CFAA context are analogous. Though online civil rights testing takes place on private websites, it is criminalized by the CFAA: By placing criminal sanctions on conduct that violates websites’ terms of service, which in turn prohibit scraping and the creation of false profiles, the government punishes testers for engaging in speech acts in violation of the First Amendment. The con­stitutional issue turns not on the location of the speech but rather on the government’s sanctioning of it. 217 See also Watchtower Bible & Tract Soc’y of N.Y., Inc. v. Village of Stratton, 536 U.S. 150, 165–66 (2002) (striking down an ordinance prohibiting canvassing on private property absent a permit because to require “a citizen [to] first inform the government of her desire to speak to her neighbors and then obtain a permit to do so” is a “dramatic departure from our national heritage and constitutional tradition”).

Protecting online civil rights testing conduct based on the theory that it is conduct incidental to speech would align with courts’ analyses of why recording and other types of newsgathering fall under the First Amendment’s purview. Not only is it immediately necessary for the final speech act, but civil rights testing, whether online or in person, also falls in line with one of the core purposes of the First Amendment: revealing the truth. 218 See Chen & Marceau, High Value Lies, supra note 104, at 1437; see also Oakes, supra note 166, at 1137 (discussing how “truth-seeking and knowledge-advancement” are values inherent in a traditional model of the First Amendment). Therefore, the interpretation of the CFAA currently used by courts, which prohibits testing activity, infringes upon protected-speech-enabling conduct.

c. Additional Barriers. — While it is likely that testing activity could be seen as conduct incidental to speech, there are further challenges to extending First Amendment protection here that should be addressed. First, as the Dietemann case illustrates, the First Amendment is not a defense to torts or crimes committed by speech actors in the course of newsgath­ering. 219 Dietemann v. Time, Inc., 449 F.2d 245, 249 (9th Cir. 1971); see also supra notes 93–95 and accompanying text (giving an overview of the Dietemann case).
In an unpublished opinion, the Ninth Circuit recently also said that the First Amend­ment does not nullify nondisclosure agreements. Nat’l Abortion Fed’n v. Ctr. for Med. Progress, 685 F. App’x 623, 626–27 (9th Cir. 2017), cert. denied, No. 17-202, 2018 WL 1568034 (U.S. Apr. 2, 2018), and No. 17-482, 2018 WL 1568035 (U.S. Apr. 2, 2018). One might argue that online civil rights testers waive any First Amendment rights they may have to con­duct testing by agreeing to abide by websites’ terms of service that proscribe this testing. However, whether a website’s terms of service are enforceable against one of the website’s users is an intensely fact-specific inquiry, depending on not only the specific terms of service but also whether the user had notice of the terms and thereby gave constructive assent to be bound by them. See Be In, Inc. v. Google, Inc., No. 12-CV-03373, 2013 WL 5568706, at *7 (N.D. Cal. Oct. 9, 2013) (noting that cases on the issue “reach disparate and fact-specific conclusions” and listing various cases). Thus, ascertaining whether, as a general matter, testers will be bound by terms of service in contract cases is highly difficult. Addi­tionally, regardless of the substantive outcome, the issue of whether a company could sue a tester for breach of contract is inapposite to the questions of whether civil rights testers engage in constitution­ally protected activity and whether criminal liability for such engagement is per­missible, and therefore such an undertaking is outside the scope of this Note.
Second, if testing activity occurs on private property, the state action requirement or public forum doctrine may affect the viability of a First Amendment claim. Third, because the scraping audit involves bots engaging in the actual data collection, one might argue that the First Amendment shouldn’t extend to scraping because bots’ activity can’t be constitutionally protected. Finally, extending protections to testers could raise issues in defining the scope of the right. However, all of these con­cerns can be adequately addressed.

While courts will likely not grant civil rights testing First Amendment protection when there are tort violations (so as not to extend newsgathering privileges to illegal conduct), 220 For an argument as to why there should be no protection in the context of tort violations, see generally Gimenez, supra note 100. this concern is not as weighty as it seems. For one, as the Seventh Circuit wrote in Desnick, testers would not be subject to trespass claims because they do not inter­fere with ownership or possession in a way that trespass law intends to protect against. 221 Desnick v. Am. Broad. Cos., 44 F.3d 1345, 1353 (7th Cir. 1995). The Ninth Circuit and Fourth Circuit have both com­mended this view, 222 Med. Lab. Mgmt. Consultants v. Am. Broad. Cos., 306 F.3d 806, 818 (9th Cir. 2002); Food Lion, Inc. v. Capital Cities/ABC, Inc., 194 F.3d 505, 517–18 (4th Cir. 1999). and other district and state courts have also suggested or adopted this idea. 223 See Access for the Disabled v. EDZ, Inc., No. 8:13-CV-3158-T-17TGW, 2016 WL 901284, at *5 (M.D. Fla. Mar. 3, 2016) (finding that tester motive does not transform the tester into a trespasser); Ouderkirk v. People for Ethical Treatment of Animals, Inc., No. 05-10111, 2007 WL 1035093, at *17 (E.D. Mich. Mar. 29, 2007) (extending Desnick’s rea­soning to the filming of a chinchilla store); Am. Transmission, Inc. v. Channel 7 of Detroit, Inc., 609 N.W.2d 607, 614 (Mich. Ct. App. 2000) (adopting Desnick’s reasoning and finding that when plaintiff investigator entered store based on misrepresentation, there was no trespass); Keyzner v. Amerlink, Ltd., 618 S.E.2d 768, 772–73 (N.C. Ct. App. 2005) (holding that plaintiff attorneys who posed as potential customers were not liable for trespass). In addition, in practice, courts have been lenient with the press in the investigative journalism context, either by not recog­nizing that a right has been violated 224 Even in Dietemann, for instance, the defendants were only liable for photographing and recording, an issue that would not be present in civil rights testing. Dietemann, 449 F.2d at 249; see also Desnick, 44 F.3d at 1353. or by giving next to nothing in the form of damages. 225 Food Lion, 194 F.3d at 522 (affirming a judgment in the amount of two dollars). Part of this may relate to the Supreme Court’s constant concern that extensive tort liability will chill speech of constitutional value. 226 See Hustler Magazine, Inc. v. Falwell, 485 U.S. 46, 52 (1988) (discussing how strict liability may chill valuable speech). While there has been pushback on this idea, see W. Watersheds Project v. Michael, No. 15-CV-00169-SWS, 2016 WL 3681441, at *5–7 (D. Wyo. July 6, 2016) (finding no First Amendment right to trespass upon private property to collect resource data), rev’d, 869 F.3d 1189 (10th Cir. 2017); Richard A. Epstein, Privacy, Publication, and the First Amendment: The Dangers of First Amendment Exceptionalism, 52 Stan. L. Rev. 1003, 1018–33 (2000), it nonetheless seems, given the case law in its favor, that the issue of trespass will not prove an obstacle in affording First Amendment protection to civil rights testing. Furthermore, in the most recent appellate case addressing the interplay between investigators and torts, the Ninth Circuit in Wasden struck down a provision criminalizing misrepresentations made to gain entry to agricultural facilities on First Amendment grounds, citing Desnick and Food Lion but making no mention of Dietemann. 227 Animal Legal Def. Fund v. Wasden, 878 F.3d 1184, 1196 (9th Cir. 2018). Given that testing involves accessing public, rather than private, domains, it seems more likely that a court would follow the Wasden court and rely on Desnick and Food Lion instead of Dietemann. 228 While Food Lion and Desnick involved public businesses, Dietemann involved entry to a home that served as a place of business. See supra section I.B.1 (discussing the factual background of these cases).

Second, with regards to the First Amendment’s state action require­ment, the Supreme Court has noted that “[t]he test is not the form in which state power has been applied but, whatever the form, whether such power has in fact been exercised.” 229 N.Y. Times Co. v. Sullivan, 376 U.S. 254, 265 (1964). Thus, even if a private litigant is seeking to enforce a tort at common law, for instance, a court’s applica­tion of that law can bring the First Amendment into play. 230 See id. Therefore, that websites promulgate terms of service preventing the occurrence of conduct incidental to speech is not a shield from First Amendment scrutiny.

The public forum doctrine may also be concerned here. The public forum doctrine is invoked when speakers seek access to public property to engage in speech acts. 231 See Cornelius v. NAACP Legal Def. & Educ. Fund, 473 U.S. 788, 800 (1985) (explaining that public forum analysis revolves around determining whether the govern­ment can limit use of its property). Thus, if websites are public forums, restric­tions on the speech acts taking place on these websites, like online civil rights testing, would implicate the First Amendment. No case has explicitly reached the question of whether public-facing websites con­stitute public forums. 232 Xiang Li, Note, Hacktivism and the First Amendment: Drawing the Line Between Cyber Protests and Crime, 27 Harv. J.L. & Tech. 301, 315 (2013). However, the Court’s recent decision in Packingham v. North Carolina 233 137 S. Ct. 1730 (2017). and the lower court rulings that have followed suggest that the public forum doctrine might encompass the internet and social media, an understanding that falls in line with what scholars have sug­gested. 234 Scholars argue that these websites should be public forums because of consent considerations, see supra note 133 and accompanying text, or because finding otherwise disservices First Amendment values, see generally Dawn C. Nunziato, The Death of the Public Forum in Cyberspace, 20 Berkeley Tech. L.J. 1115 (2005) (discussing why considering the internet to be private erodes First Amendment values). The Packingham Court, in overturning a North Carolina statute that restricted sex offenders’ access to commercial social networking websites, noted that cyberspace has become the most important place for the exchange of ideas. 235 Packingham, 137 S. Ct. at 1735. In likening social media websites to “the modern public square,” analogizing to streets or parks that were “quintessential forum[s]” for the exercise of First Amendment rights, and failing to men­tion the public–private nature of websites on the internet, 236 Id. at 1735, 1737. the Court implied that for First Amendment purposes, the internet might very well be considered a public forum. Lower courts since Packingham seem to have adopted this message, 237 See, e.g., Davison v. Loudoun Cty. Bd. of Supervisors, 267 F. Supp. 3d 702, 716 (E.D. Va. 2017) (describing Facebook as a public space and citing Packingham). In fact, Packingham was invoked in one recent case dealing with the CFAA and terms of service. In hiQ Labs, Inc. v. LinkedIn Corp., plaintiff hiQ sought access to LinkedIn after LinkedIn sent it a letter to cease and desist unauthorized data scraping in violation of LinkedIn’s terms of service. No. 17-cv-03301-EMC, 2017 WL 3473663, at *2 (N.D. Cal. Aug. 14, 2017). LinkedIn invoked the CFAA and reserved the right to pursue litigation, and hiQ subsequently filed a complaint asserting denial of access to publicly available LinkedIn profiles in violation of California’s common law, its unfair competition statute, and the California Constitution. Id. In coming to the conclusion that it “ha[d] serious doubt whether LinkedIn’s revoca­tion of permission to access the public portions of its site renders hiQ’s access ‘without authorization’ within the meaning of the CFAA,” the court cited to Packingham and noted that a “general understanding of the open nature of the Web squares with language used in” that decision. Id. at *7–8. suggesting that while the public forum doc­trine might very well be implicated in the case of online civil rights testers, it may not pose an obstacle.

A third issue that might be raised is that in the scraping audit, the tester is not conducting the protected activity at issue; rather, bots are conducting the protected activity (that is, the investigation and subsequent data collection). 238 See Sandvig et al., supra note 5, at 12–13. Nonetheless, a First Amendment right extends to the bots’ activities. Many lower courts have recently granted First Amendment protections for algorithmically produced search-engine results on the basis that the algorithms embody and perpetuate the speech of the engi­neers who create them. 239 See, e.g., Jian Zhang v. Baidu.com Inc., 10 F. Supp. 3d 433, 438–39 (S.D.N.Y. 2014). In fact, some of the technology companies suspected of algorithmic discrimination are the very entities that have been litigating to protect algorithm results. See Mark Joseph Stern, Speaking in Code: Are Google Search Results Protected by the First Amendment?, Slate (Nov. 20, 2014), http://www.slate.com/articles/technology/future_tense/2014/11/are_google_
results_free_speech_protected_by_the_first_amendment.html [http://perma.cc/9KP5-GTBZ] (noting that Google has been developing a First Amendment defense that would extend constitutional protections to algorithms since 2003).
Analogously, the bots that are created and directed by online civil rights testers should also receive First Amendment protections.

A final consideration to address is that of scope: How far could the arguments that give rise to First Amendment protections for civil rights testing reach? Could these arguments allow for protections for scraping or creation of false profiles to gather data that would be used for purposes other than civil rights enforcement? Given that many of the cases involving scraping and the CFAA deal with entities seeking confiden­tial data from their competitors, companies may have significant concerns with the scope of the right advocated for in this Note. And under a plain reading of the case law, it’s unclear whether purpose or intent would be a notable factor in deciding whether the First Amendment reaches particular data-gathering conduct. The Supreme Court has explicitly dispelled the notion that the First Amendment specially privileges jour­nalists, 240 See Branzburg v. Hayes, 408 U.S. 665, 684–86 (1972). and some of the “right to record” decisions have foundations in this very case law. 241 See, e.g., Fields v. City of Philadelphia, 862 F.3d 353, 359 (3d Cir. 2017) (noting that because “the press has [the right to record], so does the public” and citing to Branzburg, 408 U.S. at 684); Crawford v. Geiger, 996 F. Supp. 2d 603, 614 (N.D. Ohio 2014) (“The First Amendment protects not just the right of the press to gather news—it affords that right to the general public as well.” (citing Branzburg, 408 U.S. at 684)). Going even further, some of these cases have explicitly disavowed the idea that the recorder must intend to disseminate the video at the time of recording. 242 Fields, 862 F.3d at 358 (criticizing the district court’s focus on whether plaintiffs had a desire to disseminate the recordings or use them to criticize the police at the moment the recordings were made). At the same time, however, courts deciding these “right to record” and other cases addressing newsgathering conduct are at least partially motivated by the public interest served by the conduct at issue. 243 See supra notes 195–196, 203 and accompanying text. Thus, despite courts’ acknowledgment that the First Amendment protects journalists and nonjournalists alike, the fact that purpose has some ana­lytical force suggests that scraping and creation of false profiles in violation of terms of service would not be protected in just any context. Though this context does not necessarily need to be enforcement of civil rights statutes, it may still need to have some significant public interest element, thereby cabining the scope of this right.

Whether testing activity is protected as conduct incidental to speech or as false speech, it nonetheless should fall under the First Amendment’s purview as a matter of doctrine and theory. Testing, like other inves­tigative tactics, directly facilitates publication; yet unlike certain investigative methods, courts have not found civil rights testing to be a tortious act. Similarly, civil rights testing neither involves tortious misrepresentations nor gives rise to any legally cognizable harm such that testers’ false speech can be regulated. 244 For cases and literature supporting this notion, see supra notes 160–162. Furthermore, the truth-seeking aims of civil rights testing align with the First Amendment’s purpose of ensuring an informed citizenry. When courts interpret the CFAA in a way that prohibits activity based on a website’s terms of service, which can include prohibitions on certain methods of online civil rights testing, the CFAA effectively restricts constitutionally protected speech activity and falls under the First Amend­ment’s scrutiny.

III. How to Protect Testers from the CFAA: Legislation or Litigation?

As established, civil rights testing is an enormously important tool for enforcing civil rights laws, and this conduct falls under the First Amend­ment’s protection. Thus, the current interpretation suggested by courts and the Department of Justice—that online civil rights testing is unlawful when prohibited by a website’s terms of service—may be unconsti­tutional. This Part analyzes possible solutions to this problem. Section III.A looks at the suggestion that Congress amend the CFAA to exclude civil rights testing, and section III.B addresses judicial resolution of the CFAA–First Amendment tension. This Part ultimately concludes that liti­gation is the optimal route to achieve the strongest pro­tection of civil rights testing.

A. Legislatively Protecting Testers or Amending the CFAA

One solution to the problem of the CFAA’s prohibition on testing conduct is congressional legislation amending the CFAA so as to preclude courts from interpretations that proscribe civil rights testing. This appears to be an attractive solution because, as the CFAA is legislation itself, it would be simplest for Congress to overrule courts’ interpreta­tions as it has done in the past. 245 For an example in which this has happened, see, e.g., Young v. United Parcel Serv., Inc., 135 S. Ct. 1338, 1343 (2015) (noting that Congress’s “Pregnancy Discrimination Act makes clear that Title VII’s prohibition against sex discrimination applies to discrimina­tion based on pregnancy” though past precedent held otherwise). In fact, congressional inaction may be costly. See William N. Eskridge, Jr., Overruling Statutory Precedents, 76 Geo. L.J. 1361, 1366–67 (1988) (explaining that the “super-strong” presumption against overruling statutory precedents can be justified on the notion that Congress’s failure to amend statutes is tacit approval of courts’ interpretation). Additionally, to protect civil rights testers more generally, Congress could use its Commerce Clause power to affirmatively shield testers from exposure to tort liability. 246 Given the connection between antidiscrimination and economic benefits, see supra note 178, it is plausible that the Commerce Clause could serve as a basis for protecting civil rights testers. However, there has been a recent trend of reading the Commerce Clause narrowly at the Supreme Court level, see David M. Driesen, The Economic/Noneconomic Activity Distinction Under the Commerce Clause, 67 Case W. Res. L. Rev. 337, 345 (2016), which may affect the viability of this approach. This would be an alterna­tive to judicial recognition of a First Amendment right to engage in civil rights testing.

The main arguments in favor of using legislation rather than adjudica­tion 247 See, e.g., J. Harvie Wilkinson, III, Cosmic Constitutional Theory: Why Americans Are Losing Their Inalienable Right to Self-Governance 8 (2012) (arguing that current constitutional theories allow judges to impose “their personal vision of the proper good” on Americans, a practice which threatens to supplant the political, and accountable, branches of government); id. at 41 (complaining of judges “creating constitutional rights with only the slightest semblance of a textual hook”); David E. Pozen, Constitutional Bad Faith, 129 Harv. L. Rev. 885, 922 (2016) (discussing “legislating from the bench” as an example of constitutional overreach, or bad faith); William H. Rehnquist, The Notion of a Living Constitution, 54 Tex. L. Rev. 693, 700 (1976) (“[T]he Constitution does not put . . . the legislative branch or the executive branch in the position of a television quiz show contest­ant so that when . . . time has elapsed and a problem remains unsolved by them, the federal judiciary may press a buzzer and take its turn at fashioning a solution.”); see also Obergefell v. Hodges, 135 S. Ct. 2584, 2624–26 (2015) (Roberts, C.J., dissenting) (arguing in favor of using the political process to legalize gay marriage). center on the notion that there is a level of legitimacy conferred upon the decisions of a democratically elected majority that is absent in judge-made law. 248 See Obergefell, 135 S. Ct. at 2627 (Scalia, J., dissenting). Some scholars say that judges who adjudicate on mat­ters involving social rights not grounded in the Constitution are considered to be “the closest thing the United States has to a governing nobility” 249 See, e.g., Ian Millhiser, The Most Incompetent Branch, 23 Geo. Mason L. Rev. 507, 507 (2016). —antidemocratic actors whose decisions are indistinguishable from those of the Justices who decided Dred Scott and Lochner. 250 Rehnquist, supra note 247, at 704 (“To the extent that one must, however, go beyond even a generously fair reading of the language and intent of [the Constitution] in order to subsume [principles protecting individual liberties or minorities’ interests], . . . they are not really distinguishable from those espoused in Dred Scott and Lochner.”). Thus, in the context of the CFAA, it could be argued that Congress is best suited to protect civil rights testers as a matter of legitimacy, for courts should not “find” new rights in the Constitution not readily apparent from the text or past precedent. 251 See Millhiser, supra note 249, at 507.

There are certain weaknesses in the argument that demo­cratic political means should be the primary vehicle for social change. For one, the above criticism rests on the assumption that there is no textual basis for the right at stake. 252 See Moore v. City of East Cleveland, 431 U.S. 494, 544 (1977) (White, J., dissenting) (“The Judiciary, including this Court, is the most vulnerable and comes nearest to illegiti­macy when it deals with judge-made constitutional law having little or no cognizable roots in the language or even the design of the Constitution.”); see also Millhiser, supra note 249, at 507 (arguing that “a judge’s legitimacy flows from a written text,” and that a “judge who divorces their opinions from a controlling text strays far afield from their constitu­tional role”); supra note 247. In the case at hand, however, as the analysis from section II.B establishes, a First Amendment right to test for discrim­ination can be predicated upon an ample supply of doctrinal founda­tions. In other words, a tester’s rights can be firmly grounded in the Constitution’s text and past precedents. Furthermore, on a more con­ceptual level, although interpreting open-ended constitutional provi­sions can seem like legislating from the bench because there is little text to guide judges, this is not necessarily Lochnerism. 253 See Michael R. Dimino, Sr., Image Is Everything: Politics, Umpiring, and the Judicial Myth, 39 Harv. J.L. & Pub. Pol’y 397, 398 (2016). Lochner has faced criticism for two reasons. One reason is that the Court second-guessed the legislature on matters solely within the legislature’s competence; the second reason similarly criticizes the Court for second-guessing the legislature, but more so because this second-guessing was done pursu­ant to “invented rights.” See Jamal Greene, The Anticanon, 125 Harv. L. Rev. 379, 418–19 (2011). To say a judicial decision is a product of Lochnerism (that is, an application of the Lochner philosophy) is on its own “enough to damn it.” John Hart Ely, The Wages of Crying Wolf: A Comment on Roe v. Wade, 82 Yale L.J. 920, 940 (1973).

Another issue with the political process argument is that it seems to ignore the current political reality of extreme polarization and the resulting barriers to creating substantive legislative change. 254 See Lee Drutman, American Politics Has Reached Peak Polarization, Vox (Mar. 24, 2016), http://www.vox.com/polyarchy/2016/3/24/11298808/american-politics-peak-polarization [http://perma.cc/D2Y7-5EU4]; see also Pozen, supra note 247, at 931 (discussing bad faith allegations between Congressional Republicans and Democrats). Constitutional law scholars have argued that although the judiciary espouses a general skep­ticism toward innovation used to circumvent this polarization, such skep­ticism has no constitutional basis, and judges should take into account political reality when adjudicating consti­tutional cases. 255 See, e.g., Gillian E. Metzger, Appointments, Innovation, and the Judicial–Political Divide, 64 Duke L.J. 1607, 1610–11 (2015). Thus, while it intuitively seems that Congress is best suited for protecting civil rights testers, given the political reality of modern times, it may be that the courts are in fact the most reliable source of protection. Though barriers exist to a successful First Amendment chal­lenge to protect testing via litigation, there are benefits to doing so that are not available with legislative remedies.

B. Litigating for First Amendment Protection

As the analysis in section II.B highlights, there are multiple lines of case law that can support First Amendment protection for civil rights testers. Thus, courts could interpret the CFAA in a way that avoids implicating the First Amendment altogether pursuant to the canon of constitutional avoid­ance, a canon which essentially states that courts should avoid ruling on constitutional issues when there is a way to resolve the issue on a nonconstitutional basis. 256 See Spector Motor Serv., Inc. v. McLaughlin, 323 U.S. 101, 105 (1944) (“If there is one doctrine more deeply rooted than any other in the process of constitutional adjudi­cation, it is that we ought not to pass on questions of constitutionality . . . unless such ad­judication is unavoidable.”).

Currently courts use website terms of service to determine when an individual has engaged in “unauthorized” access because they consider a violation of terms of service to be the proper metric for finding a CFAA violation. 257 See, e.g., EF Cultural Travel BV v. Zefer Corp., 318 F.3d 58, 62–63 (1st Cir. 2003) (noting that lack of authorization can be established by explicit statements on websites re­stricting access, such as terms of service defining what acts are unauthorized or forbidden). Courts could instead attempt to redefine such an interpretation when applied to testers, perhaps by analogizing to courts’ discussion of testers and torts 258 See supra notes 93–99 and accompanying text (discussing lower courts’ treatment of testers and tort liabilities). because the CFAA was meant to apply to the online equivalent of “trespass.” 259 See supra note 136 (discussing the legislative intent behind the CFAA). Because courts have dismissed the notion that civil rights testers can incur tort liability for going onto private property and misrepresenting themselves as potential patrons, 260 See supra notes 220–226 and accompanying text (discussing courts’ analyses re­garding testing and torts). courts analogously could dismiss actions against testers who violate websites’ terms of service (and thereby commit computer trespass) to conduct testing. Under the canon of constitutional avoidance, because courts could inter­pret the CFAA in a way that does not implicate the First Amendment, they should.

Were an opportunity to establish affirmative constitutional protection for civil rights testers available, however, it should be pursued. 261 As some scholars have noted, courts’ assessment of when avoidance is possible is “plausibly (and reasonably) affected by the perceived practical stakes,” Ryan D. Doerfler, High-Stakes Interpretation, 116 Mich. L. Rev. 523, 552 (2018), so a court may nonetheless address this issue. Though the viability of arguments in favor of tester protection may be tempered by the relative incoherence of First Amendment doctrine, 262 See Post, Recuperating, supra note 73, at 1249–50. the novelty of the arguments themselves, and the Supreme Court’s fairly deferential application of strict scrutiny in at least some First Amend­ment cases, 263 See Dimino, supra note 253, at 405 (arguing that the Court in Williams-Yulee v. Florida Bar, 135 S. Ct. 1656, 1669 (2015), abandoned narrow tailoring in its scrutiny of a state law that prevented judges from personally soliciting campaign contributions). there are appreciable benefits to mounting a challenge to the CFAA on First Amendment grounds, so long as it is possible.

For example, there is a certain sense of permanence that comes from constitutional interpretation in that only the Supreme Court is able to reinterpret the substantive rights found in the Constitution 264 See Marbury v. Madison, 5 U.S. 137, 177 (1803). For an example of when the Supreme Court has overruled itself in interpreting the Constitution, see W. Coast Hotel Co. v. Parrish, 300 U.S. 379, 400 (1937) (overruling Adkins v. Children’s Hosp., 261 U.S. 525 (1923)). —absent an amendment, which is notoriously difficult to pass. 265 See Rehnquist, supra note 247, at 705 (noting that a legislator might “seek to run the more difficult gauntlet of amending the Constitution”). In addition, though a statutorily created right has a sense of legitimacy in that it derives from the authority of a democratically elected Congress, a consti­tutional right tends to be perceived, whether accurately or not, as even more legitimate and respectable because of the extremely venerable posi­tion the Constitution holds. 266 See Richard H. Fallon, Jr., Legitimacy and the Constitution, 118 Harv. L. Rev. 1787, 1827–34 (2005). But see 3 Bruce Ackerman, We the People: The Civil Rights Revolution 34 (2014) (arguing that civil rights statutes have constitutional status).

While the desire to use the legislature to enact social change is com­pelling, litigating for First Amendment protection would be the best strategy to protect civil rights testers as there are far more chal­lenges to amending the CFAA or separately creating an affirmative statutory protec­tion for testers. Litigation allows for a sense of permanence and legitimacy that does not necessarily come from general congressional legislation, and though the argument is novel, it is nonetheless worth making. As a matter of doctrine and policy, civil rights testing can fit under certain aspects of the First Amendment, and recognizing that would strengthen testers’ abil­ity to engage in the truth-revealing work that they do while also protecting information-gathering tactics that crucially facilitate protected speech acts.

Conclusion

The First Amendment, with its protections for speech, expression, and the press, has been widely considered a hallmark of democracy. In order for it to serve its purposes, the First Amendment must be inter­preted broadly. 267 Richmond Newspapers, Inc. v. Virginia, 448 U.S. 555, 576 (1980) (“For the First Amendment does not speak equivocally. . . . It must be taken as a command of the broadest scope that explicit language, read in the context of a liberty-loving society, will allow.” (in­ternal quotation marks omitted) (quoting Bridges v. California, 314 U.S. 252, 263 (1941))). And to that end, testing activity, as a method of investi­gating discriminatory practices, should be protected by the Constitution. The role of testers, especially civil rights testers, in uncovering legal vio­lations is of the utmost importance to achieving the aims of a right to free speech and press: “‘A critical, independent, and investigative press is the lifeblood of any democracy.’” 268 Sue Valentine, Mandela’s Legacy of Media Freedom Stands Its Ground, in Comm. to Protect Journalists, Attacks on the Press: Journalism on the World’s Front Lines 133, 136 (2014 ed.) (quoting Nelson Mandela).

Whether intentional or not, algorithms can function in discrimina­tory ways. Such discrimination can violate civil rights laws and carry great consequences for society. The CFAA, based on courts’ current interpreta­tion, penalizes activities prohibited by websites’ terms of service and effectively blocks researchers, journalists, and civil rights advocates from conducting online civil rights testing. In doing so, the CFAA comes into tension with the First Amendment, which can be read to protect civil rights testing. To remedy this problem, rather than amending the CFAA or cre­ating a separate statutory protection for civil rights testers, courts should acknowledge that the First Amendment does protect testing conduct and invalidate the current interpretation of the CFAA that pro­hibits online civil rights testing.