As evidenced by the confusion expressed by multiple Justices in last Term’s Gonzalez v. Google, there is little consensus as to the scope of Section 230, the law that broadly immunizes internet platforms from liability for third-party content. This is particularly striking given that no statute has had a bigger impact on the internet than Section 230, often called the “Magna Carta of the internet.”

In this essay I argue that Section 230, despite its simple-seeming language, is a deeply ambiguous statute. This ambiguity stems from a repeated series of errors committed by Congress, the lower courts, and the Supreme Court in the drafting, enactment, and early judicial interpretation of the statute.

This diagnosis, which I lay out in Part I, sets the stage for Part II, in which I consider three potential paths forward for the judicial interpretation of Section 230. In particular, I focus on a novel interpretative approach, by which courts would interpret Section 230 immunity narrowly in order to spur large technology companies to lobby Congress to act, thereby forcing Congress to clarify the scope of platform intermediary liability. But this approach carries substantial risks of disrupting the internet in the time between the judicial reinterpretation of Section 230 and Congress’s response, and thus represents at best an imperfect solution to the legislative and judicial mistakes that attended Section 230’s origins.

Introduction

No statute has had a bigger impact on the internet than Section 230. By preventing any online platform from being “treated as the publisher or speaker” of third-party content,¹1.47 U.S.C. § 230(c)(1) (2018). it has enabled the business models of the technology giants that dominate the digital public sphere. Whether one champions Section 230 as the “‘Magna Carta’ of the internet”²2.Noa Yachot, The “Magna Carta” of Cyberspace Turns 20: An Interview With the ACLU Lawyer Who Helped Save the Internet, ACLU (June 23, 2017), https://www.aclu.org/news/free-speech/magna-carta-cyberspace-turns-20-interview-aclu-lawyer-who-helped (https://perma.cc/LAR9-N73Z). or vilifies it as the “law that ruined the internet,”³3.Steve Randy Waldman, The 1996 Law That Ruined the Internet, Atlantic (Jan. 3, 2021), https://www.theatlantic.com/ideas/archive/2021/01/trump-fighting-section-230-wrong-reason/617497 (https://perma.cc/7SZT-6PQQ). there is no doubting that it “made Silicon Valley.”⁴4.Anupam Chander, How Law Made Silicon Valley, 63 Emory L.J. 639, 650-57 (2014).

Thus, it is remarkable that—nearly 30 years after its enactment—basic questions about its meaning and scope remain unanswered. Consider the oral argument in last Supreme Court Term’s Gonzalez v. Google, in which the Court was set to decide whether Section 230 immunizes platforms for the act of recommending third-party content to users. This question is of immense practical importance to platforms and is hardly an esoteric corner case of platform intermediary liability.

Yet multiple justices expressed uncertainty, even bewilderment, over how to apply Section 230 to this central issue: Justice Thomas was “confused,” Justice Jackson was “thoroughly confused,” and Justice Alito was “completely confused.”⁵5.Transcript of Oral Argument at 72, 64, 34, Gonzalez v. Google LLC, 598 U.S. 617 (2023) (No. 21-1333) (hereinafter Gonzalez transcript). Justice Kagan, in the most memorable portion of argument, quipped, “We really don’t know about these things. You know, these are not like the nine greatest experts on the Internet.”⁶6.Id. at 46. Given the tone of oral argument, it was unsurprising when, several months later, the Court punted, resolving Gonzalez on unrelated grounds and, in a tacit admission to how difficult the problem was, never reaching the Section 230 issue.⁷7.Gonzalez, 598 U.S. 617; see also Scott R. Anderson, Quinta Jurecic, Alan Z. Rozenshtein & Benjamin Wittes, The Supreme Court Punts on Section 230, Lawfare (May 19, 2023, 12:00 PM), https://www.lawfaremedia.org/article/the-supreme-court-punts-on-section-230 (https://perma.cc/E945-Z8SE). The muddled, out-of-place discussion of Section 230 in this Term’s NetChoice oral arguments further demonstrates a lack of consensus on Section 230’s core features.⁸8.See Jeff Kosseff, Have Trouble Understanding Section 230? Don’t Worry. So Does the Supreme Court. Lawfare (Mar. 7, 2024), https://www.lawfaremedia.org/article/have-trouble-understanding-section-230-don-t-worry.-so-does-the-supreme-court (https://perma.cc/K4X4-YA58) (describing the many “misinterpretations (about Section 230 that) were on full display during the NetChoice oral arguments”).

Why has the Supreme Court failed so dramatically to provide basic clarity to such an important law? And what will happen the next time the Court is called to interpret it? In this Article, I argue that Section 230, despite its simple-seeming language, is a deeply ambiguous statute and that this ambiguity stems from a repeated series of errors committed by Congress, the lower courts, and the Supreme Court in the drafting, enactment, and early judicial interpretation of the statute.⁹9.In this Article, I limit my analysis to that part of Section 230 that protects platforms for content that they host. In particular, I do not address the separate question of to what extent Section 230 protects platforms for their decisions to remove content. See generally Adam Candeub & Eugene Volokh, Interpreting 47 U.S.C. § 230(c)(2), 1 J. Free Speech L. 175 (2021). This diagnosis, which I lay out in Part I, sets the stage for Part II, in which I consider three potential paths forward for the judicial interpretation of Section 230. In particular, I focus on a novel interpretative approach, by which courts would interpret Section 230 immunity narrowly in order to spur large technology companies to lobby Congress to act, thereby forcing Congress to clarify the scope of platform intermediary liability. Admittedly, this approach carries substantial risks of disrupting the internet in the time between the judicial reinterpretation of Section 230 and Congress’s response. My conclusion is that, unfortunately, there is no perfect solution to the legislative and judicial mistakes that attended Section 230’s origins.

My hope is that my argument adds a new perspective to the voluminous scholarly literature on Section 230. Scholars debating how Section 230 should be interpreted often focus on the underlying policy question: the appropriate scope of intermediary liability immunity.¹⁰10. A notable exception to this trend is Jeff Kosseff’s work on the history of Section 230. See Jeff Kosseff, The Twenty-Six Words that Created the Internet (2019), at chs. 2-3; Jeff Kosseff, A User’s Guide to Section 230, and a Legislator’s Guide to Amending It (or Not), 37 Berkeley Tech. L.J. 757, 761-73 (2022); and Jeff Kosseff, What Was the Purpose of Section 230? That’s a Tough Question., 103 B.U. L. Rev. 763 (2023). I rely on Kosseff’s work throughout this Article. This is understandable, but it threatens to obscure the fact that the task of interpreting Section 230 is just that, a problem first and foremost of statutory interpretation, in which traditional principles of faithful agency to congressional intent ought to play a leading role. This Article is an attempt to interpret Section 230 from statutory interpretation first principles.

I. The Ambiguities of Section 230

The key provision of Section 230—the “twenty-six words that created the internet,” in the words of its leading historian Jeff Kosseff—is (c)(1): “No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider.”¹¹11. 47 U.S.C. § 230(c)(1) (2018).

In most Section 230 cases, which involve a person suing an online platform or service for harm caused by third-party content, the key question is what it means for the platform or service to be “treated as the publisher or speaker” of third-party content. On one extreme, this language can be interpreted very broadly, so as to prohibit virtually all lawsuits against platforms for harm involving third-party conduct. On the other extreme, the language can be interpreted very narrowly, permitting platform liability in a variety of contexts, such as when the platform knowingly hosts harmful third-party content or affirmatively recommends or promotes such content on its service. With some notable exceptions, courts have read Section 230 expansively.

To see why the language of Section 230 is ambiguous as to its scope—or, at minimum, that the broad interpretation is not unambiguously correct—it is important to situate Section 230 within its broader legal and policy context. Two historical facts are particularly important: the pre–Section 230 caselaw that spurred its drafting and the broader legislative package of which Section 230 was just one part.

A. The Judicial Context of Section 230

Section 230 arose because of idiosyncrasies in how courts applied the common law of distributor liability to defamation claims against online platforms. At the time of Section 230’s enactment, liability for transmission of defamatory third-party content depended on the nature of the transmission. As described in the Restatement (Second) of Torts, a “publisher” of such content could be held liable as long as the publisher acted negligently, even without knowledge that the content was defamatory.¹²12. 3 Restatement (Second) of Torts § 558 (Am. L. Inst. 1977).

By contrast, someone who merely “delivers or transmits defamatory matter published by a third person”—in other words, a “distributor”—would only be liable if they knew or had reason to know that the content was defamatory. Thus, while bookstores or libraries did not need to review every book they offered in advance to avoid liability for selling or circulating a defamatory book, they could be liable as distributors if they circulated books already known to them to be defamatory. Similarly, a telegraph operator was not liable for transmitting a message that the operator did not know (or had reason to know) was libelous.¹³13. Id. § 581(1) cmts. e-f.

The question for courts applying distributor liability to internet platforms was whether the platforms were publishers of their third-party content—and so could be held negligently liable for it—or whether they were merely distributors—and so could only be held liable if they knew or had reason to know about the defamatory nature of the material. Two cases, both decided in the early 1990s, addressed this question and established the emerging law to which Section 230 was a reaction.

In the first case, Cubby, Inc. v. CompuServe, Inc., the owner of a media news outlet sued CompuServe, one of the main early online service providers (along with America Online (AOL) and Prodigy), because CompuServe hosted a forum in which allegedly defamatory material was posted about the news outlet. The court held that CompuServe was not liable for the defamatory material because CompuServe was a distributor, not a publisher, of the forum and thus could only be held liable if it knew that the forum it hosted was transmitting defamatory content. In holding that CompuServe was a distributor and not a publisher, the court emphasized that CompuServe did not review content before it was released on the forum and made available across CompuServe. Thus, in the court’s view, “CompuServe has no more editorial control over such a publication than does a public library, book store, or newsstand, and it would be no more feasible for CompuServe to examine every publication it carries for potentially defamatory statements than it would be for any other distributor to do so.”¹⁴14. Cubby, Inc. v. CompuServe, Inc., 776 F. Supp. 135, 140 (S.D.N.Y. 1991).

The second case, which was decided in 1995 and led directly to Section 230’s introduction into Congress a year later, was Stratton Oakmont, Inc. v. Prodigy. Stratton Oakmont, the securities firm led by the soon-to-be disgraced Jordan Belfort, sued Prodigy for allegedly defamatory comments made about the firm on Prodigy’s finance-related “Money Talk” message board. Unlike CompuServe, which did not moderate its forums, Prodigy moderated the Money Talk board in a variety of ways, and, as the court wrote, “held itself out as an online service that exercised editorial control over the content of messages posted on its computer bulletin boards, thereby expressly differentiating itself from its competition and expressly likening itself to a newspaper.”¹⁵15. Stratton Oakmont, Inc. v. Prodigy Services Co., No. 31063/94, 1995 WL 323710, at *2 (N.Y. Sup. Ct. May 24, 1995). On this basis, and explicitly distinguishing Cubby, the court held that Prodigy should be subject to publisher, not merely distributor, liability for the allegedly defamatory content.

From the beginning, it was clear that Stratton Oakmont perversely incentivized platforms not to moderate content, since it was Prodigy’s decision to moderate some content that led the court to hold it liable as a publisher for any content it allowed to remain on its platform.¹⁶16. Indeed, the judge in Stratton Oakmont recognized this, though he argued that the economic benefits that a platform would gain from moderating its content and thus being more user- and family-friendly would outweigh the additional litigation risk. Id. at *5. For this reason, the decision, despite arising from a state trial court, received national attention. In stories published the day after the case was decided, the New York Times described AOL’s general counsel as “hop[ing] that on-line services would not be forced to choose between monitoring bulletin boards and assuming liability for users’ messages,”¹⁷17. Peter H. Lewis, Judge Allows Libel Lawsuit Against Prodigy to Proceed, N.Y. Times, May 26, 1995, at D4. and Time noted that Prodigy was “[i]ronically” being held more liable for its users’ speech than were other, non-moderated, services.¹⁸18. Netwatch . . . Unease After Prodigy Ruling, Time (May 26, 1995), https://content.time.com/time/nation/article/0,8599,3939,00.html (https://perma.cc/KU65-MACE).

It is indisputable that Section 230 was written in large part to overturn Stratton Oakmont. Christopher Cox and Ron Wyden, Section 230’s sponsors, have confirmed this in recent years.¹⁹19. See Kosseff, supra note 10, at 60-61. The question that has bedeviled interpreters of Section 230 ever since is how much beyond Stratton Oakmont it was meant to go—specifically, what it means for a platform to be “treated as the publisher or speaker of any information provided by another.” Indeed, Section 230 goes farther than is strictly required to overrule Stratton Oakmont, since it prohibits treating platforms as publishers outright, rather than only on the basis of the platforms’ moderation practices (the issue in Stratton Oakmont). And Cox and Wyden have stated that they intended Section 230 to go further than simply overturning Stratton Oakmont.²⁰20. See Jeff Kosseff, The Lawsuit Against America Online That Set Up Today’s Internet Battles, Slate (July 14, 2020, 5:45 AM), https://slate.com/technology/2020/07/section-230-america-online-ken-zeran.html (https://perma.cc/RB2G-8C2A) (“Did (Fourth Circuit Judge J. Harvie Wilkinson) misinterpret Section 230 when he ruled against Zeran? Both of its authors, Cox and Wyden, told me that Wilkinson got it right.”); see also Chris Cox & Ron Wyden, Reply Comments on NTIA Petition for Rulemaking to Clarify Provisions of Section 230 of the Communications Act of 1934, at 15-17 (Sept. 17, 2020), https://www.fcc.gov/ecfs/document/10917190303687/1 (https://perma.cc/XS8G-T4KK) (arguing that Section 230 does not permit the use of negligence standards).

But Section 230’s text cannot be considered in isolation, nor can the views of its sponsors be treated as conclusive evidence of Congress’s intent. To understand Congress’s intent behind Section 230, it is necessary to turn to the law of which it was only a small part: the Communications Decency Act of 1996.

B. The Legislative Context of Section 230

When Section 230 was enacted in 1996, it was as part of a broader congressional response to the perceived dangers of the internet—specifically, the problem of children being exposed to inappropriate content, especially pornography. In fact, Section 230, although it has come to assume a central role in internet law, was only one small, and relatively obscure, part of a much broader legislative package, the Communications Decency Act (CDA) of 1996, itself part of the Telecommunications Act of 1996.²¹21. Although the law is commonly described as “Section 230 of the Communications Decency Act,” it was actually Section 509 of the Telecommunications Act of 1996, of which Title V (covering sections 501-09) was the Communications Decency Act. Section 509 created a new section 230 in the Communications Act of 1934, codified at 47 U.S.C. § 230. Thus, the proper name of Section 230 should be “Section 230 of the Communications Act of 1934, as amended” or “Section 509 of the Telecommunications Act of 1996.” Jeff Kosseff, What’s in a Name? Quite a Bit, If You’re Talking About Section 230, Lawfare (Dec. 19, 2019, 1:28 PM), https://www.lawfaremedia.org/article/whats-name-quite-bit-if-youre-talking-about-section-230 (https://perma.cc/4FVQ-A6VX). To avoid confusion, I will simply refer to the law as Section 230.

The CDA, introduced by Senator James Exon, criminalized the knowing transmission of “obscene” or “indecent” messages to minors. Internet advocates were concerned that the combination of criminal penalties and vague, broad language would cripple the then-nascent internet by causing platforms to censor large quantities of content so as not to risk violating the CDA.

As an alternative, Christopher Cox and Ron Wyden, then both members of the House of Representatives, introduced alternative language in their “Internet Freedom and Family Empowerment Act,”²²22. H.R. 1978, 104th Cong. (1995). which sought to encourage platforms to moderate content on a voluntary, rather than mandatory, basis. This text would become Section 230.

The first two sections of Section 230 set out various findings and policy statements and illustrate Section 230’s multiple—and potentially conflicting—goals. Section 230 was intended to accomplish the goal of protecting children but through a different mechanism, the removal of “disincentives for the development and utilization of blocking and filtering technologies that empower parents to restrict their children’s access to objectionable or inappropriate online material.”²³23. 47 U.S.C. § 230(b)(4) (2018).

But Cox and Wyden also had loftier free expression goals in mind. Recognizing that the internet represented an “extraordinary advance in the availability of educational and informational resources” and “a forum for a true diversity of political discourse, unique opportunities for cultural development, and myriad avenues for intellectual activity,” Cox and Wyden sought “to promote the continued development of the Internet,” in particular the “vibrant and competitive free market that presently exists for the Internet . . . unfettered by Federal or State regulation.”²⁴24. Id. at § 230(a), (b).

After the House and Senate passed their respective versions of the Telecommunications Act of 1996—the House including Cox and Wyden’s Internet Freedom and Family Empowerment Act and the Senate including Exon’s CDA—the bills went to the conference committee. For reasons that remain unclear, the conference committee, rather than taking the logical step of choosing between the Exon and Cox-Wyden proposals, included both provisions as part of a single “Communications Decency Act,” with the Cox-Wyden proposal as an added final section to Exon’s original legislation.

The conference committee report devoted only half a page to describing Section 230 and focused entirely on its effect of “protect[ing] from civil liability those providers and users of interactive computer services for actions to restrict or to enable restriction of access to objectionable online material.” It specifically listed overruling Stratton Oakmont as “[o]ne of the specific purposes” of Section 230. It argued that, by treating “providers and users as publishers or speakers of content that is not their own because they have restricted access to objectionable material,” the decision created “serious obstacles to the important federal policy of empowering parents to determine the content of communications their children receive through interactive computer services.”²⁵25. H.R. Rep. No. 104-458, at 194 (1996) (Conf. Rep.).

Shortly after the reconciled Telecommunications Act was enacted into law, civil liberties groups led by the ACLU challenged the CDA in court. Specifically, they alleged that the criminal penalties (Exon’s original CDA) violated the First Amendment. The Supreme Court agreed in Reno v. ACLU,²⁶26. 521 U.S. 844 (1997). leaving Section 230 as the only remaining operative provision of the CDA.

The importance of this convoluted drafting history is this: to understand the intent of Congress in enacting Section 230, it is not enough to look at Section 230 in isolation. Cox and Wyden might have preferred for Section 230 to fully replace Exon’s original CDA, but Congress chose to enact both bills. Whether the conference committee was intentional or merely sloppy in combining two dramatically different and arguably inconsistent approaches to platform liability, its choice commits courts to interpret both provisions such that “effect is given to all . . . provisions” of the ultimately enacted CDA so that “no part will be inoperative or superfluous, void or insignificant.”²⁷27. Hibbs v. Winn, 542 U.S. 88, 101 (2004).

The competing purposes behind the Communications Decency Act demonstrate how ambiguous its provisions really are. Specifically, it is a mistake, as some courts have done, to single out the promotion of “freedom of speech in the new and burgeoning Internet medium” as Congress’s overriding purpose in enacting Section 230.²⁸28. Zeran v. AOL, Inc., 129 F.3d 327, 330 (4th Cir. 1997); see also Jane Doe No. 1 v. Backpage.com, LLC, 817 F.3d 12, 29 (1st Cir. 2016) (discussing the “First Amendment values that drive” Section 230). Section 230, after all, was not enacted as part of the “Internet Freedom and Family Empowerment Act,” a title that, in emphasizing platform and user control, accurately reflects Cox and Wyden’s personal intentions. Rather, it became law as part of the “Communications Decency Act,” which was, as Danielle Citron and Benjamin Wittes have observed, “by no stretch of imagination a libertarian enactment.”²⁹29. Danielle Keats Citron & Benjamin Wittes, The Internet Will Not Break: Denying Bad Samaritans § 230 Immunity, 86 Fordham L. Rev. 401, 404 (2017). This suggests that a central if not primary goal was to encourage the removal of “indecent” content online.

Even recognizing the censorious purposes of the CDA, defenders of a broad reading of Section 230 immunity might nevertheless argue that the CDA, taken as a whole, represented a deal by which platforms would be held criminally liable for indecent content in exchange for receiving civil immunity. This is an intriguing suggestion, but it suffers from two flaws. First, there is no evidence in either the text or the legislative history of this deal. Second, and more fundamentally, if this interpretation of Section 230 is correct, there is a strong argument that the Supreme Court erred in not striking down Section 230 alongside the rest of the CDA in Reno: by severing Section 230 from the rest of the statute, the Court undermined congressional intent by destabilizing the compromise that the CDA created.³⁰30. See, e.g., Murphy v. Nat’l. Collegiate Athletic Ass’n, 584 U.S. 453, 481 (2018) (holding that severability is unavailable where it is “evident that Congress would not have enacted those provisions which are within its power, independently of those which are not”) (quoting Alaska Airlines, Inc. v. Brock, 480 U.S. 678, 684 (1987)) (cleaned up).

C. Distributor vs. Publisher Liability

The obvious first question to ask is whether treating a platform as a distributor falls within Section 230’s prohibition against treating platforms as publishers. As noted above, Congress could have overturned Stratton Oakmont by making clear that the mere fact of moderating some content would not by itself permit the platform to be “treated as a publisher” of content it failed to moderate. But Congress went further, prohibiting platforms from ever being treated as publishers of third-party content. The question is what Congress meant by this additional degree of protection.³¹31. There is an additional interpretive option: that (c)(1) was only ever meant as a definitional, rather than immunity-granting, provision, and the only immunity granted by Section 230 was that of (c)(2), which immunizes platforms when they remove certain categories of content. See, e.g., Doe v. GTE Corp., 347 F.3d 655, 660 (7th Cir. 2003) (Easterbrook, J.); Shlomo Klapper, Reading Section 230, 70 Buff. L. Rev. 1237, 1281 (2022). My argument does not rely on this claim, so I do not address it.

This was the issue in the first major case interpreting Section 230, Zeran v. AOL, decided by the Fourth Circuit in 1997. The case is the most important judicial decision on Section 230 to date, and its holding that Section 230 applies broadly to any attempt to hold a platform liable for third-party content remains canonical. Nevertheless, the background legal and legislative context described above demonstrate that Zeran was based on flawed reasoning that failed to recognize Section 230’s ambiguity.

Zeran involved Kenneth Zeran, an ordinary person whose life was thrown into chaos after a series of messages on an AOL bulletin board falsely connected him to the Oklahoma City terrorist bombing. Zeran repeatedly informed AOL about this content and asked AOL to take down the offending messages, but AOL refused. Zeran sued, arguing that AOL was negligent in not responding adequately to the false messages about him. But because Section 230 had just been enacted, Zeran could not argue that AOL was liable in the way a publisher would be—negligently, whether they knew about the offending messages or not. So Zeran instead argued that AOL was liable as a distributor—that is, because AOL knew about the messages (because Zeran kept telling them).

The court rejected Zeran’s argument that Section 230 should be read narrowly, as applying only to publisher liability. First, it argued that distributor liability was a subset of publisher liability, and thus Section 230’s prohibition on treating a platform as a publisher of third-party content extended to attempts to treat a platform as a distributor of that content. To support this argument, the court primarily relied on an influential contemporary tort law treatise, which used a broad definition of “publication” to describe any act of transmission of defamatory material, including by distributors (which the treatise described as “secondary publishers”). Thus, the fact that newspapers could be held negligently liable for transmitting defamatory content while bookstores could only be liable if they had knowledge was merely a difference in liability between different types of publishers.³²32. Id. at 332-33 (citing W. Page Keeton et al., Prosser and Keeton on the Law of Torts § 113 (5th ed. 1984)).

Second, citing to Section 230’s findings and policy statements, the court argued that Congress’s purpose in enacting Section 230 was to counter “the threat that tort-based lawsuits pose to freedom of speech in the new and burgeoning Internet medium.”³³33. Id. at 330. Because the “specter of tort liability in an area of such prolific speech would have an obvious chilling effect,” the court reasoned, “Congress considered the weight of the speech interests implicated and chose to immunize service providers to avoid any such restrictive effect.”³⁴34. Id. at 331. And because distributor liability would expose platforms to “potential liability each time they receive notice of a potentially defamatory statement,” platforms would have “a natural incentive simply to remove messages upon notification, whether the contents were defamatory or not.” Thus, “liability upon notice has a chilling effect on the freedom of Internet speech.”³⁵35. Id. at 333.

The problem with both of those arguments is that they overlook important features of Section 230’s legal and legislative history. First, even assuming that the common law at the time of Section 230’s enactment treated distribution as a subset of publication, Section 230 was enacted as a direct response to the Stratton Oakmont decision, which treated distribution not as a subset of publication, but rather as a distinct category.³⁶36. Stratton Oakmont, Inc. v. Prodigy Services Co., No. 31063/94, 1995 WL 323710, at *3 (N.Y. Sup. Ct. May 24, 1995). Although legal terms are presumed to carry their established common law meanings when they are used in legislation, this presumption can be overridden when there is evidence of contrary legislative intent. Thus, as a matter of congressional intent, it is at least ambiguous as to whether the 1996 Congress, rather than just Cox and Wyden,³⁷37. See supra note 22. meant to include distribution within the scope of Section 230’s immunity provision.³⁸38. See, e.g., David R. Sheridan, Zeran v. AOL and the Effect of Section 230 of the Communications Decency Act upon Liability for Defamation on the Internet, 61 Albany L. Rev. 147, 168-69 (1997).

Second, putting Section 230 in its proper legislative context—as part of the broader Communications Decency Act—demonstrates that Congress’s overall purpose could not possibly have been to “keep government interference in the medium to a minimum.”³⁹39. Zeran, 129 F.3d at 330. To the contrary, imputing such a purpose to Congress would lead to a distorted understanding of Congress’s intent. It’s clear, as the court argued, that notice-based liability would likely result in platforms taking down large swaths of legitimate content. But it’s hardly clear that the same Congress that voted for criminal liability for transmitting obscene material to children—with all of such liability’s predictable chilling effects—would have suddenly objected to those chilling effects when they resulted from tort liability.

Even the court’s more limited claim that Section 230 represented a “policy choice . . . not to deter harmful online speech through the separate route of imposing tort liability on companies that serve as intermediaries for other parties’ potentially injurious messages” is questionable.⁴⁰40. Id. at 330-31. As Justice Thomas has noted, because the criminal-law provisions of the Communications Decency Act included civil enforcement for the “knowing . . . display” of indecent material to children—that is, distributor liability—it is unlikely that Section 230 was intended to eliminate such liability altogether.⁴¹41. Malwarebytes, Inc. v. Enigma Software Group USA, LLC, 141 S. Ct. 13, 15 (2020) (Thomas, J., statement respecting denial of certiorari).

To be sure, there is an argument for reading Section 230 as eliminating distributor liability for platforms that also respects Congress’s intent to encourage platforms to aggressively moderate content. Specifically, the Zeran court was right to recognize the possibility that “notice-based liability would deter service providers from regulating the dissemination of offensive material over their own services,” since “any efforts by a service provider to investigate and screen material posted on its service would only lead to notice of potentially defamatory material more frequently and thereby create a stronger basis for liability.”⁴²42. Zeran, 129 F.3d at 333.

The question is whether that incentive would be stronger than the incentive to remove content. There are two reasons to doubt that it would be, at least in many circumstances. First, under a distributor-liability regime platforms would still be liable for content that they were informed was illegal even if they had not proactively sought to detect such content. Second, and more importantly, a completely unmoderated platform would drive away consumers (which is why the major platforms spend hundreds of millions on proactive moderation). Even with the litigation risk that proactive moderation would create, platforms would likely moderate content anyway so as to create an environment that users wanted to be in.

Admittedly, this incentive to moderate is greater for large rather than small platforms, since the former have more to gain financially and have the resources to spend on extensive moderation. A narrow reading of Section 230 would thus favor large over small platforms, many of which will likely shut down rather than face the increased liability risk that a narrow interpretation of Section 230 would create. This is unlikely to be good for free expression on the internet. But given the importance that Congress placed on encouraging moderation and preventing harmful content in enacting the Communications Decency Act, an interpretation of Section 230 that favors large over small platforms is not inconsistent with congressional intent.

The point of the above argument is not to establish definitively that Section 230 was not intended to eliminate distributor liability (though I believe the weight of evidence does support that conclusion). Rather, the point is to establish a case that Section 230 is at the very least ambiguous as to whether it eliminated distributor liability.

D. Liability for Algorithmic Amplification

But the question of whether Section 230 eliminates distributor liability has not been a live issue in the courts for decades. At both the state and federal levels, courts have adopted Zeran’s broad interpretation of the statute. More recent legal debates around Section 230 have instead focused on whether platforms are liable when they affirmatively promote illegal content, for example through algorithmic recommendations.⁴³43. The question, in other words, is, if Zeran is correct, is there a situation in which recommendations are nevertheless unprotected by Section 230? If Zeran is wrong and Section 230 is narrowly about eliminating negligence liability for platforms, then the answer is easier: recommendations are unprotected at least when the platform knew or had reason to know that they were recommending illegal content.

This was the core question in the Gonzalez case. In November 2015, a series of attacks by the Islamic State killed more than 130 people in Paris, including Nohemi Gonzalez, a 23-year-old California college student who was on a foreign exchange program. Gonzalez’s family sued Google under a provision of federal law that imposes liability on anyone who aids, intentionally or not, a terrorist attack. The family argued that Google was liable because it promoted ISIS content on its YouTube platform to users by means of recommendation algorithms, thereby aiding in ISIS’s recruitment efforts and thus aiding, even if indirectly, ISIS in carrying out the 2015 attacks. The district and circuit courts held that Section 230 barred Gonzalez’s lawsuit, and, in early 2023, the Supreme Court heard oral argument on the issue. (A related case, Twitter v. Taamneh, addressed whether, absent Section 230, the platforms would be substantively liable under the federal anti-terrorism statutes.⁴⁴44. 598 U.S. 471 (2023). )

The case hinged on what it means for a legal claim to “treat[]” a platform “as the publisher” of third-party content. The Ninth Circuit decision in Gonzalez, as well as other circuits coming to a similar conclusion, interpreted this provision broadly, using the everyday meaning of publisher: one who engages in the many activities associated with publishing, which include promotion and recommendation.⁴⁵45. Gonzalez v. Google LLC, 2 F.4th 871, 892 (9th Cir. 2021); see also Force v. Facebook, Inc., 934 F.3d 53, 65 (2d Cir. 2019). By contrast, Gonzalez and the government characterized Section 230 as a limited intervention in defamation law. They argued for a narrow construction of the provision that focused on the nature of publisher liability under common law: the transmission of a communication whose content is defamatory (or otherwise tortious). On this reading, Section 230 should not bar suits that, rather than treating the defendant as liable simply for retransmitting the tortious communication, seek to hold the defendant liable for harms that go beyond mere retransmission—such as, for personalized recommendations.⁴⁶46. Brief for Petitioners at 19-26, Gonzalez v. Google LLC, 598 U.S. 617 (2023) (No. 21-1333; Brief for the United States as Amicus Curiae in Support of Vacatur at 13-17, Gonzalez, 598 U.S. 617 (No. 21-1333).

This distinction is a subtle one, and it is hard to say that either side had a knock-down argument regarding Section 230’s textual meaning. Nor does legislative history provide much insight. Although recommendation algorithms were not unknown when Section 230 was enacted in 1996,⁴⁷47. See, e.g., Nicholas J. Belkin & W. Bruce Croft, Information Filtering and Information Retrieval: Two Sides of the Same Coin?, 35 Commc’ns ACM 29, 29 (Dec. 1, 1992). they played nothing like the central role that they do today. Thus, it is unsurprising that neither the text of Section 230 nor the legislative history addresses the question of liability for recommendation algorithms. As Justice Kagan noted at oral argument, “everybody is trying their best to figure out how [Section 230] . . . which was a pre-algorithm statute[,] applies in a post-algorithm world.”⁴⁸48. Gonzalez transcript, supra note 5, at 9.

Ultimately, the Court did not decide the Section 230 issue in Gonzalez, holding instead that the plaintiff’s substantive tort claim was faulty.⁴⁹49. Taamneh, 598 U.S. at 478. But the Section 230 question is not going away. Algorithmic recommendations are at the heart of modern social media, and courts, including the Supreme Court, will no doubt continue to grapple with whether Section 230 protects platforms for such recommendations.

As with the issue of distributor versus publisher liability, the best argument as a matter of statutory interpretation for a broad reading of Section 230 liability immunity in the context of algorithmic recommendation is that it would effectuate one of the primary goals of the statute: encouraging platforms to moderate content and provide tools for users to moderate that content. Algorithmic moderation—including content removal, downranking, and “shadowbanning”—is increasingly the key mechanism by which platforms moderate objectionable content.⁵⁰50. For an overview of how platforms can respond to objectionable content, see Eric Goldman, Content Moderation Remedies, 28 Mich. Tech. L. Rev. 1, 23-40 (2021). If platforms are held liable for personalized recommendations, they may decide not to do any automated screening or ranking at all, out of an abundance of caution against incurring liability.

But the reason this argument should not have been decisive in Zeran is also why it should not be decisive here. If algorithms really are necessary for platforms to avoid becoming cesspools of offensive content, then platforms will continue to use them no matter the litigation risk (if only to the extent necessary to maintain users and preserve advertising revenue). As noted above, this will likely lead to some smaller platforms leaving the market entirely. But it is inconceivable that Facebook or Twitter would choose to shutdown rather than invest the necessary resources in content moderation. The question is whether platforms will, under a liability regime that permits platforms to be sued for the harms caused by their algorithmic recommendations, host on net more or less harmful content. The impossibility of answering this question is yet another reason why Section 230 is, properly interpreted, ambiguous as to the question of liability for algorithmic amplification.

II. How to Interpret An Ambiguous Statute

In the previous section, I argued that Section 230 is profoundly ambiguous as to congressional intent when it comes to the major questions of contemporary internet liability. In this section I consider the three ways that courts could respond to this ambiguity: (1) interpret Section 230 based on their own view of what would lead to the socially best outcome; (2) continue the status quo interpretation and hope that Congress clarifies the statute; or (3) interpret the statute narrowly so as to encourage Congress to act. I focus the bulk of my analysis on option three because, unlike options one or two, it has not previously been explored, either by courts or scholars. Like all the options, it has its drawbacks, but it also has substantial advantages and should be seriously considered as a path out of the Section 230 morass.

To emphasize a point made in the Introduction,⁵¹51. See supra note 10 and accompanying text. my argument differs from the existing literature on Section 230, in which commentators have largely focused on the policy merits and demerits of competing interpretations of the statute. My aim is not to determine the best policy outcome as a matter of free expression on the one hand or the prevention of harm on the other. Indeed, I am undecided on the question of what the proper intermediary liability regime is. Rather than focus on what the socially optimal answer is, my approach, in classic legal-process tradition, is about who should decide—specifically, how the courts can help Congress best engage in this important issue of national policy.⁵²52. See Daniel B. Rodriguez, The Substance of the New Legal Process, 77 Calif. L. Rev. 919, 950 (1989) (noting that the “comparative institutional competence argument is a familiar one” within the legal process tradition).

A. Policy-Based Interpretation

When faced with an ambiguous statute—either because the text is unclear, Congress was trying to satisfy multiple competing goals, or the statute is applied to novel situations that were unforeseen by the drafters—courts are naturally tempted to pivot away from legislative intent and instead interpret the statute so as to lead to the best policy result. Given Section 230’s many ambiguities, it is thus unsurprising that debates over its interpretation often focus on the possible outcomes.

For example, in the Gonzalez oral argument, Justice Kavanaugh worried that a narrow reading of Section 230 could lead to “nonstop” lawsuits that would “create a lot of economic dislocation” and “really crash the digital economy.”⁵³53. Gonzalez transcript, supra note 5, at 54, 81. Likewise, Chief Justice Roberts suggested that such lawsuits might cause the internet to “be sunk.”⁵⁴54. Id. at 79. On the other hand, an overly broad reading of Section 230 could permit serious harms. For example, Justice Sotomayor worried about giving platforms carte blanche to use “an algorithm that inherently discriminates against people.”⁵⁵55. Id. at 42.

But courts are institutionally ill-suited to conduct this kind of policy-based reasoning. Such analysis requires a great deal of empirical research, which courts have neither the expertise nor the capacity to undertake. More fundamentally, policy judgment requires making contestable value-based choices for which courts lack the necessary democratic legitimacy.

The debate over platform intermediary liability exemplifies these two points. First, knowing what the “right” intermediary liability regime is requires knowing the effects of any particular intermediary liability regime. Admittedly, no institution is in a particularly good position to know this, but at least Congress (or, more likely, the administrative agency that Congress might delegate regulatory power to) can in principle make a more informed, empirically driven conclusion.

Second, any liability regime requires trading off important values—at a first approximation, free expression on the one hand, and harm to individuals on the other hand.⁵⁶56. In reality, it is more complicated, in both directions. Internet users whose speech is restricted because of a narrow immunity regime may suffer harms in addition to expressive ones, as the FOSTA example, infra, demonstrates. Similarly, internet users who are harmed by content that platforms host or promote under a wide immunity regime will may feel less free to express themselves. There is no right answer to this question, and, in a democratic society, these sorts of contestable value choices are properly made by the more democratic institutions of government.

B. Status Quo Interpretation

For the reasons described above, it would be much preferable for Congress, rather than the courts, to decide the difficult policy issues and value tradeoffs that are at the center of the debate over intermediary liability for platforms. But what should courts do in the meantime, while they wait for Congress to clarify the meaning of Section 230?

One approach is for courts to do nothing—that is, maintain the legal status quo. There are two justifications for this approach—one that is specific to Section 230, and the other that is more generally applicable to statutory interpretation.

The argument for maintaining the Zeran approach to Section 230 is that the modern internet—in particular the giant platforms that dominate internet usage—has come to rely on it even in the absence of a Supreme Court opinion explicitly endorsing that approach, despite decades of opportunities for the Court to do so.⁵⁷57. See, e.g., Doe v. AOL , 783 So. 2d 1010 (Fl. 2001), cert. denied, 534 U.S. 891 (2001); Ben Ezra, Weinstein, & Co. v. AOL, 206 F.3d 980 (10th Cir. 2000), cert. denied, 531 U.S. 824 (2000); Batzel v. Smith, 333 F.3d 1018 (9th Cir. 2003), cert. denied, 541 U.S. 1085 (2004). As I have previously argued, “where trillions of dollars of economic value depend on the lower court decisions in question, the Supreme Court should be sensitive to the real-world effects of overturning case law that has been in place for decades.”⁵⁸58. Alan Z. Rozenshtein, Section 230 and the Supreme Court: Is Too Late Worse Than Never?, Lawfare (Oct. 20, 2020, 1:01 PM), https://www.lawfaremedia.org/article/section-230-and-supreme-court-is-too-late-worse-than-never (https://perma.cc/F9GM-UNMD).

The issue is not just that limiting Section 230 would force the large platforms to revise their business models. It would also create enormous uncertainty in the law as to what the platforms could be held liable for. As Blake Reid has noted, “Section 230 effectively absolved judges of having to decide how non-internet-specific bodies of law—contract, property, tort, and the like—would apply to Internet platforms in most circumstances involving the carriage or moderation of user-generated content.” This “interpretative debt” means that suddenly adopting a more limited conception of Section 230 would “implicate[] not only a huge volume of difficult, resource-intensive, and complex legal work needed to bring a vast range of laws up to speed in a new context, but also the consequences of disrupting surrounding social, cultural, political, democratic, and economic contexts of platforms that have evolved for the entire life of the commercial Internet in reliance on an essentially laissez-faire treatment of carriage and moderation.”⁵⁹59. Id. (manuscript at 15).

The other reason for courts, including the Supreme Court, to uphold the Zeran interpretation of Section 230 is if they think that strictly upholding statutory stare decisis—that is, upholding a previous statutory interpretation even in the face of compelling arguments that the interpretation was incorrect—might encourage Congress to step in and clarify the statute at issue by “articulating a clear and unyielding division of responsibility.” ⁶⁰60. Lawrence C. Marshall, “Let Congress Do It”: The Case for an Absolute Rule of Statutory Stare Decisis, 88 Mich. L. Rev. 177, 210 (1989). If Congress wants the law to change, it will have to change it.

This approach was suggested by several justices in the Gonzalez oral argument. Justice Kagan suggested that restricting the broad reach of Section 230 is “something for Congress to do, not the Court[.]”⁶¹61. Gonzalez transcript, supra note 5, at 46. And Justice Kavanaugh doubted that the Court, rather than Congress, was the “right body to draw back from” a broad reading of Section 230.⁶²62. Id. at 54. Instead, he argued that it would be “better for [the Supreme Court] to keep [Section 230] the way it is” and “to put the burden on Congress to change that” so that Congress could “consider the implications and make these predictive judgments[.]”⁶³63. Id. at 82.

If Congress does choose to step in and override or otherwise clarify the meaning of Section 230, its doing so against the broad Zeran interpretation of the statute has the benefit of allowing gradual reform. For example, in 2018, Congress enacted the Allow States and Victims to Fight Online Sex Trafficking Act (FOSTA), carving out sex trafficking content from Section 230’s protections.⁶⁴64. Allow States and Victims to Fight Online Sex Trafficking Act of 2017, 47 U.S.C. § 230(e)(5) (2018). As Reid notes, this will force courts to work through the difficult legal issues that Section 230 has heretofore obviated, but in a narrow sandbox, limiting disruption on the broader internet. If the courts preserve the Zeran approach, Congress could respond by enacting more FOSTA-like laws, gradually experimenting with different intermediary liability regimes in a way that would be less disruptive than if the Court were to radically cut back Section 230’s protections for platforms.⁶⁵65. Reid, supra note 62 (manuscript at 23-26).

Both arguments for the status quo—the one from reliance and the one from spurring congressional involvement—are plausible, but they suffer from important flaws. First, the reliance interest that platforms have in a broad reading of Section 230 has a flipside in the ongoing harms that many users experience from that broad reading. For courts to assume that the platforms’ reliance is more important than users’ ongoing harm is for judges to engage in precisely the same sort of unwarranted policy balancing that accompanies policy-based interpretation.

Second, unlike most high-profile cases implicating statutory stare decisis, Section 230 has never been authoritatively interpreted by the Supreme Court. For the Supreme Court to treat the existing jurisprudential consensus as precedent, it would in effect allow lower court opinions to be binding on it, reversing the usual structure of the federal courts and committing the country to a consequential jurisprudential status quo that has not benefited from review by the highest court.

Third, the assumption that maintaining the statutory status quo will encourage congressional action may itself be incorrect. As the next section explains, the best chance to get Congress to clarify what it wants internet intermediary liability to look like might require not preserving the Section 230 status quo, but rather dramatically altering it.

C. Preference-Eliciting Interpretation

Maintaining the interpretative status quo may sometimes cause Congress to act. But it is just as likely to inhibit congressional action, given the reality of legislative inertia. Legislative agendas are busy, and enacting legislation is hard. This means that the interpretative status quo will encourage, rather than discourage, Congress to act only when the power of those groups in society—and thus their congressional allies—that are disadvantaged by the status quo outweighs the power of those groups that are benefited by the status quo.

More generally, an interpretation that benefits a politically strong group is less likely to lead to a congressional response than is an interpretation that benefits a politically weak group. In the former scenario, the politically powerful group benefits from maintaining the status quo; in the latter, the group benefits from changing the status quo. In both cases, the powerful group can, through lobbying and other forms of political influence, shape the congressional agenda to its benefit. This means that the interpretation that is most likely to spur congressional action will generally be the interpretation that is least favorable to the most influential interest groups, whether or not the interpretation is the status quo.

Building on this insight, Einer Elhauge has developed a theory of what he calls “preference-eliciting default rules.” Elhauge defines an approach to statutory interpretation by which courts “choose the interpretations that are most likely to elicit legislative reactions, which will produce a statutory result that embodies enactable preferences more accurately than any judicial estimate could.”⁶⁶66. Einer Elhauge, Statutory Default Rules: How to Interpret Unclear Legislation 152 (2008).

Elhauge’s preference-eliciting approach has many moving parts, but the aspect that is most relevant to how courts should interpret Section 230 is the presumption that ambiguous statutes should be interpreted in such a way as to disfavor the interests of the more politically powerful group. This is not because powerful interest groups are “bad” in either a moral or social sense, but because they are best placed to spur Congress to action when judicial interpretation disfavors their interests.⁶⁷67. Importantly, Elhauge demonstrates that such an approach to statutory interpretation is not only normatively desirable but also is compatible with existing practice, even though courts generally are not explicit about the preference-eliciting justification. For example, Elhague notes that the rule of lenity, which requires courts to interpret ambiguous criminal statutes in favor of the defendant, can be justified on the theory that it disfavors law enforcement, which has substantially more political power than the opposing group, criminal defendants, and is thus well placed to lobby the legislature to expand criminal laws that have been interpreted narrowly. Id. at 168-81. By contrast, courts tend to interpret antitrust and tax laws broadly, which accords with Elhauge’s observation that the groups that benefit from the narrow interpretation of those laws are well placed to spur Congress to action when judicial decisions cut against their interests. Id. at 181, 361 n.33; Nat’l Gerimedical Hosp. & Gerontology Ctr. v. Blue Cross of Kansas City, 452 U.S. 378, 388-89 (1981) (antitrust); United States v. Wells Fargo Bank, 485 U.S. 351, 354 (1988) (tax).

What would a narrow interpretation of Section 230 look like? In short, it would look like the reverse of the key Section 230 cases. It would reverse Zeran and find that Section 230 does not eliminate distributor liability. It would permit product-liability lawsuits against companies that knowingly design systems that are easily abused to cause harm.⁶⁸68. See, e.g., Herrick v. Grindr LLC, 765 F. App’x 586 (2d Cir. 2019). And, in cases like Gonzalez v. Google, it would permit lawsuits specifically alleging that personalized algorithmic recommendation had increased the harm of third-party content. To be clear, in none of these cases would the platform necessarily lose on the underlying legal claim. Rather, the platform would simply no longer be able to claim automatic immunity and would instead have to defend itself under the applicable substantive law of negligence, defamation, products liability, etc.

From the justification of the preference-eliciting approach—that it encourages Congress to clarify its intent—Elhauge derives three conditions that are required for its appropriate use. First, it only applies when the underlying statute is actually ambiguous as to both the initial and current Congresses’ intent(s). If the statute clearly meant to help the politically powerful interest group, or if it is clear that this is what the current Congress would prefer, then that is the interpretation that should be adopted. Second, the preference-eliciting approach should only be used when it will substantially and meaningfully increase the chance that Congress will actually respond by expressing its intention through legislation or active deliberation. Third, the costs—in terms of the interpretation’s effects on public policy—of the interim period between the judicial interpretation and the legislative clarification must be acceptable. For example, if encouraging Congress to clarify its intentions requires tanking the economy, it is probably not worth it.⁶⁹69. Elhauge, supra note 70, at 155-67. The rest of this section applies each of these three conditions to consider the case for a narrow interpretation of Section 230.

1. Congressional Intent

With respect to the first condition—that Section 230 does not capture Congress’ intent—Part I has explored at length why the Zeran approach is at best ambiguous as to, if not outright in conflict with, Congress’s intent when it enacted Section 230. But what about the current Congress? As Elhauge notes, if a statute can be interpreted in such a way as to bring it in line with Congress’s current enactable political preferences, courts should adopt that interpretation.

In the case of Section 230, it is at best unclear what the current Congress’s enactable political preference is on the issue of platform liability for third-party content. It is possible that if Congress seriously put the issue on the legislative agenda, it would reaffirm Zeran’s broad reading of Section 230, or even go further and overturn the scattered cases that have found Section 230 inapplicable. But it is just as likely that Congress would narrow immunity. After all, Congress did just that in 2018 when, by overwhelming margins in both the House and Senate, it carved content violating sex trafficking law out of Section 230’s immunity provision.⁷⁰70. One might argue that because Congress enacted FOSTA without overturning the existing lower-court consensus on Section 230’s broad scope, it in effect “ratified” that interpretation. See, e.g., Lorillard v. Pons, 434 U.S. 575, 580 (1978) (“Congress is presumed to be aware of an administrative or judicial interpretation of a statute and to adopt that interpretation when it re-enacts a statute without change.”). The problem with applying that logic to FOSTA is that FOTSA was not a comprehensive replacement for Section 230, but rather a small alteration to its substantive scope. There is no reason to think that the legislative coalition that pushed FOSTA through either intended or thought it possible to reevaluate the fundamental structure of Section 230. And the current spate of Section 230 reform proposals suggests that there is substantial appetite for further limitations.⁷¹71. See Meghan Anand et al., All the Ways Congress Wants to Change Section 230, Slate (Sept. 19, 2023), https://slate.com/technology/2021/03/section-230-reform-legislative-tracker.html (https://perma.cc/QA35-ZBMC).

2. Likelihood of Congressional Action

Supporting the second requirement for a preference-eliciting interpretation—that it meaningfully increases the probability that Congress will engage with the underlying legal and policy issue—are two observations. First, Section 230 operates as a valuable subsidy to technology platforms. Second, these platforms form a powerful interest group.

On the first point, both the supporters and critics of Section 230 agree that it has been critical to the development of the modern internet and, more specifically, to the business models of the technology giants. Section 230’s immunity shield is effectively a subsidy to platforms—especially the largest ones—that no comparable sector of the communications sector enjoys. Without Section 230, platforms would either be open to massive amount of litigation cost or would have to sharply curtail how much user content they host, thus leading to potentially lower user “engagement.” Either way, they would pay a substantial financial cost and would be incentivized to lobby Congress.

But would they succeed? In other words, are tech companies good lobbyists? There are plenty of good reasons to think so. Technology companies are some of the largest lobbyists, with three companies—Amazon, Meta (Facebook’s parent company), and Alphabet (Google’s parent company)—in the top 20 of all spenders on lobbying.⁷²72. Top Spenders, OpenSecrets, https://www.opensecrets.org/federal-lobbying/top-spenders (https://perma.cc/X5YA-SH7S) (last visited Apr. 2, 2024). Apple is trying to catch up, having increased its lobbying budget by more than 40% in 2022.⁷³73. Lauren Feiner, Apple Ramped Up Lobbying Spending in 2022, Outpacing Tech Peers, CNBC (Jan. 23, 2023, 11:26 AM EST), https://www.cnbc.com/2023/01/23/apple-ramped-up-lobbying-spending-in-2022-outpacing-tech-peers.html (https://perma.cc/D8VA-FYA4). Overall, the big five tech companies—Amazon, Meta, Alphabet, Apple, and Microsoft—spent nearly $70 million in federal lobbying, an increase from 2021.⁷⁴74. Id.

And the result of all that lobbying has been impressive, especially when it comes to blocking legislation. Minnesota Senator Amy Klobuchar has publicly complained that Congress has done “zilch” on passing privacy or transparency legislation for the tech industry because “there are lobbyists around every single corner of this building that have been hired by the tech industry.”⁷⁵75. Cat Zakrzewski, Tech Companies Spent Almost $70 million Lobbying Washington in 2021 as Congress Sought to rein in Their Power, Wash. Post (Jan. 21, 2022, 2:51 PM EST), https://www.washingtonpost.com/technology/2022/01/21/tech-lobbying-in-washington (https://perma.cc/KCE6-DUY2). The tech industry’s ability to stymie legislation goes beyond these areas. Its highest-profile success was a successful Google-led campaign against the controversial “Stop Online Piracy Act” (SOPA), which would have strengthened copyright enforcement. It has also been a strong lobby against legislative efforts that would mandate law-enforcement access to encrypted data.⁷⁶76. Matt Apuzzo et al., Apple and Other Tech Companies Tangle With U.S. Over Access to Data, N.Y. Times (Sept. 7, 2015), https://www.nytimes.com/2015/09/08/us/politics/apple-and-other-tech-companies-tangle-with-us-over-access-to-data.html (https://perma.cc/J2K6-M5XT).

Just as importantly, the tech industry can help not only defeat legislation, but also get it enacted. For example, it was a driving force behind the CLOUD Act, which created a process by which foreign governments can directly serve process on U.S. platforms for user data. It also successfully lobbied to be able, as part of the FREEDOM Act, to provide more public information on U.S. government requests for user data.⁷⁷77. Taylor Hatmaker, As the CLOUD Act Sneaks into the Omnibus, Big Tech Butts Heads with Privacy Advocates, TechCrunch (Mar. 22, 2018, 8:06 PM EDT), https://techcrunch.com/2018/03/22/cloud-act-omnibus-bill-house (https://perma.cc/G7SM-7XQT).

To be sure, the technology industry’s record in enacting its policy preferences has not been one of untrammeled success. The Kids Online Safety Act, which would impose substantial child-protection obligations on platforms, is gaining momentum in Congress.⁷⁸78. Natasha Singer, Bipartisan Bill Aims to Protect Children Online, N.Y. Times (Jan. 31, 2024), https://www.nytimes.com/2024/01/31/technology/congress-social-media-safety.html (https://perma.cc/7VW3-36J2). And outside Washington, the tech companies have experienced defeats—for example, with regard to California’s state data privacy law (which the tech industry continues to lobby against),⁷⁹79. Issie Lapowsky, Tech Lobbyists Push to Defang California’s Landmark Privacy Law, Wired (Apr. 19, 2019, 3:09 PM), https://www.wired.com/story/california-privacy-law-tech-lobby-bills-weaken (https://perma.cc/BKZ5-NLR2). or, in particular, with respect to European regulation, where the Digital Services Act and the Digital Markets Act are poised to radically reshape who the tech industry can do business.

But it remains the case that, in Congress at least (where any response to a narrow interpretation of Section 230 would occur), the tech industry has substantial agenda-setting power. This agenda-setting power need not be such as to guarantee a congressional response to justify a narrow, preference-eliciting reading of Section 230—a high-enough likelihood is enough. Nor is it necessary that the result of tech lobbying be an outcome that is favorable to the tech companies. As long as they feel that congressional action is more likely than not to be better for them than a narrow judicial interpretation of Section 230, they will be incentivized to lobby Congress to take up the issue. Recent history suggests that they have a meaningful chance of success.

What about the last time Congress amended Section 230—what might that tell us about the political economy of Section 230 reform? After all, when Congress enacted FOSTA, carving out sex trafficking content from Section 230’s protections, it did so not because courts had interpreted Section 230 narrowly and the platforms had complained, but precisely for the opposite reason. The courts had interpreted Section 230 broadly so as to immunize platform hosting of sex-trafficking content, and the platforms were perceived as not doing enough to change their own behavior.⁸⁰80. See Doe v. Backpage.com, LLC, 817 F.3d 12 (1st Cir. 2016); Kendra Albert et al., FOSTA in Legal Context, 52 Colum. Hum. Rts. L. Rev. 1084 (2021). Does this show that Congress can act on issues of platform intermediary liability in the absence of a demand from platforms? If so, that would be an argument against a narrow preference-eliciting approach, not because such an approach wouldn’t elicit congressional action, but because such an approach would not be necessary to elicit such action.

The answer is that FOSTA, although clearly a legislative defeat for the platforms, was nevertheless a minor one. Despite all the attention rightfully paid to it and the profound effects it had on the sex-work community, it never fundamentally threatened the big tech companies. Indeed, Meta ultimately came out in favor of it. In addition, FOSTA imposed comparatively greater burdens on small companies than on large ones, which could absorb the increased liability. But this is because FOSTA increased liability at the margin. Had FOSTA been broad enough to fundamentally threaten the business models of the incumbent platforms, they would have fought much harder against it (as they did against SOPA). Similarly, while there may be future FOSTA-scale carveouts to Section 230 even while the Zeran approach continues to dominate, a wholesale revision of Section 230 is unlikely.

3. The Benefits of the Narrow Approach Versus the Costs

The final consideration for using the preference-eliciting approach is whether the benefits of congressional clarification outweigh whatever policy costs would result from the preference-eliciting interpretation.

As I have suggested throughout, the main benefit of congressional clarification as to the scope of platform intermediary liability would be that the balance of the competing values at stake would better reflect democratic preferences rather than judicial fiat. This by itself is enough for a prima facie case supporting a preference-eliciting interpretation of Section 230.

One might grant that a narrow interpretation of Section 230 would lead large technology platforms to lobby Congress to enact a successor statute but nevertheless worry that such lobbying would be done in narrow interest of large tech companies, not smaller platforms or users as a whole. For example, it might actually be better for incumbents like Facebook or Twitter if Congress’s replacement statute is a bit stingier with liability immunity than what the current judicial interpretation of Section 230 provides. This is because large incumbent platforms may be able to better absorb litigation costs and liability risk than small upstarts. In this way, a liability regime that is somewhere between the generous status quo and the limited regime that would result from the preference-eliciting approach might be optimal for technology giants, even if it is suboptimal for other platforms or users as a whole.

This is a fair concern, to which there are two responses. The first is to defend the large platforms as a reasonable, even if not ideal, proxy for public preferences, since their financial incentive is closely tied to their ability to satisfy their hundreds of millions of American users. The second response is to note that the concern about the distorting effects of interest group lobbying is not so much a critique of preference-eliciting interpretation but rather a much broader complaint about the legislative process itself. The preference-eliciting approach is justified not by a rosy view of Congress’s ability to legislate perfectly, but by a hard-nosed realization that, in a mature, diverse democracy, there is no better available institution than Congress, aided by the regulatory state, to determine contested issues of social and economic policy.

There is no escaping the role of powerful interest groups in democratic lawmaking. Indeed, interest-group politics is democratic politics to the extent that interest-group pressure is a key driver of congressional behavior. This may be far from ideal, but it is unclear what the better alternative would be or why insulating the current interpretation of Section 230—which cannot be justified in the face of Section 230’s ambiguities—from the democratic process is either democratically legitimate or socially optimal.

Congressional action would have additional benefits. In particular, Congress, unlike the courts, can regulate on a clean slate. It need not, as courts must when interpreting Section 230, confine itself to addressing the problem of platform liability through the blunt instrument of all-or-nothing liability immunity as determined by courts. Rather, it could consider different models—for example, a notice-and-takedown system modeled on the Digital Millenium Copyright Act (DMCA), which allows copyright holders to request the removal of infringing content from websites and provides those websites with “safe harbor” from liability if the websites comply with the DMCA’s requirements.⁸¹81. 17 U.S.C. § 512(c)(1) (2018). Indeed, a DMCA-like notice-and-takedown system has become the most commonly cited alternative model for platform intermediary liability, both in the United States⁸²82. See, e.g., Allison Tungate, Bare Necessities: The Argument for a ‘Revenge Porn’ Exception in Section 230 Immunity, 23 Info. & Commc’ns Tech. L. 172, 178-80 (2014); Vanessa S. Browne-Barbour, Losing Their License to Libel: Revisiting § 230 Immunity, 30 Berkeley Tech. L.J. 1505, 1546-50 (2015); Ann Bartow, Online Harassment, Profit Seeking, and Section 230, 95 B.U. L. Rev. Annex 101, 102 (2015). and in other countries,⁸³83. See, e.g., Law Comm’n of Ontario, Defamation Law in the Internet Age 13-14 (2020). and bipartisan legislation has been introduced to bring notice and takedown to Section 230.⁸⁴84. See, e.g., Internet Platform Accountability and Consumer Transparency Act, S. 483, 118th Cong. (2023).

While the DMCA has been legitimately criticized as permitting abusive takedown notices of non-infringing content,⁸⁵85. See, e.g., Brent Skorup & Jennifer Huddleston, The Erosion of Publisher Liability in American Law, Section 230, and the Future of Online Curation, 72 Okla. L. Rev. 668-70 (2020). notice-and-takedown systems can be calibrated at different levels in terms of what a platform has to do to get safe-harbor protection.⁸⁶86. For example, the European Union’s Electronic Commerce Directive provides the basis for member states to set up their own notice-and-takedown safe harbor regimes for online content, but it provides flexibility for each state to determine the details of those regimes. Council Directive 2000/31/EC, 2000 O.J. (L 178), art. 14. Thus, Congress, either in legislation or through delegation to an administrative agency, could set out safe-harbor standards that balanced, albeit imperfectly, the incentives of platforms to moderate content against the incentives to keep it up.

Nor is a notice-and-takedown regime the only feasible alternative. One option would be to use the regime for trademark infringement as a model, which combines complete immunity for “innocent” platforms but allows plaintiffs to get injunctions ordering the platforms to remove harmful content.⁸⁷87. Mark A. Lemley, Rationalizing Internet Safe Harbors, 6 J. on Telecomm. & High Tech. L. 101, 115 (2007). Another option would be for Congress to delegate authority to an administrative agency, such as the Federal Communications Commission, to set out requirements that platforms would have to meet to get liability protection.⁸⁸88. In 2020, the Trump administration began a regulatory process that would have had the FCC to interpret the scope of Section 230, but the FCC’s jurisdiction over Section 230 was never clearly established. See Valerie C. Brannon & Eric N. Holmes, Cong. Rsch. Serv., R46751, Section 230: An Overview 36-42 (2021). Limiting the scope of Section 230 would open up more room for experimentation by state regulators, who are currently limited by Section 230’s preemption provision.⁸⁹89. 47 U.S.C. § 230(e)(3) (2018). Cf. Jack Goldsmith & Eugene Volokh, State Regulation of Online Behavior: The Dormant Commerce Clause and Geolocation, 101 Tex. L. Rev. 1083 (2023) (suggesting that geolocation technology would often allow state regulation of internet entities under the Dormant Commerce Clause). And more generally, encouraging Congress to reconsider third-party liability rules would offer an opportunity to update those rules for emerging technologies like generative AI.

Ultimately, no alternative to Section 230 will be without cost or controversy, but such is the reality given the diverse set of views in society as to how platforms ought to moderate content. It is premature to assume that a Section 230-like regime is the best we can do.

The main cost of a narrow interpretation of Section 230 is in how platforms will respond to limit their potential liability, especially if courts strike at the heart of the judicial interpretation of Section 230—Zeran’s elimination of distributor liability. Platforms would likely respond to such a decision by massively increasing moderation, removing not just illegal content but also large amounts of legal content, just to be on the safe side. This is not a hypothetical concern—it is exactly what platforms did after FOSTA, when some platforms disappeared altogether and the ones that remained removed not just sex-trafficking content but also anything having to do with sex work. The result was not merely less expression writ large, but also an environment that in some respects was more dangerous for sex workers, who could no longer use the internet to communicate with each other and screen potentially dangerous clients.⁹⁰90. See generally Danielle Keats Citron & Quinta Jurecic, FOSTA’s Mess, 26 Va. J. L. & Tech. 1 (2023).

Platforms could respond to a narrow reading of Section 230 in other ways that would worsen the user experience. For example, if Section 230 is interpreted as not protecting algorithmic amplification, companies could respond by replacing algorithmic recommendations with chronological newsfeeds, which—especially for high-volume platforms—might make it less useful for users.

In some cases, a narrow reading of Section 230 might even work to exacerbate the very problems that critics of Section 230 are trying to solve. For example, if Section 230 is interpreted as only immunizing platforms for publisher, rather than distributor, liability—that is, only for content that they do not know to be illegal—platforms might respond by doing less proactive investigation of content, so as to deny knowledge of the harmful content that they post (though, as noted above, this incentive is limited by the platforms’ need to provide users with good experiences).

Whether these costs—exacerbated, as already noted, by the platforms’ immense reliance interest on the current judicial interpretation of Section 230—outweigh the benefits of encouraging congressional clarification depends on a number of factors that are, unfortunately, impossible to pin down for certain. First, we cannot know the extent to which platforms will react in the ways described above; this depends on their litigation risk tolerance. As noted above, there is no question that many platforms, especially smaller ones, will exit the market.

However, this concern, while serious, should not be overstated. There is no reason to think that all of them would do so—certainly not the largest platforms, who make hundreds of billions of dollars a year and who can afford (just like companies in other industries) “cost of doing business” litigation. American platforms operate globally, including in many jurisdictions which lack anything like Section 230’s liability protections. This is evidence that, while Section 230 may well have been of existential importance to the internet when it was enacted, its role today is more limited, however important or beneficial it remains.

Second, the total costs of a narrow interpretation depend on how Congress reacts to a narrow judicial interpretation and how it responds substantively in crafting a replacement statute. If Congress acts quickly and chooses to explicitly enact the sorts of protections that the courts have read into Section 230, the adjustment costs will be lower. But the longer Congress waits, and the stingier it is with liability immunity, the more platforms will restrict online speech. And it is possible that even a full-court press by the technology industry will prove insufficient to get Congress, with its well-known disfunctions, to act.

Third, and most fundamentally, the decision to put the above effects on the “cost” side of the ledger begs the important question of what is a cost versus a benefit. For example, if one thinks that much of the content on the internet is harmful, they may judge the benefits of increased moderation to outweigh the costs to expression (especially if they think that one of the effects of harmful content is to intimidate others from exercising their own expressive rights). Further, if one believes that current judicial interpretations of Section 230 are out of whack with current enactable congressional preferences, then the benefits of a narrow interpretation are not only substantive but, to the extent that it will lead to congressional action, democratic as well.

There is no avoiding the fact that the debate over Section 230 is fundamentally a debate about competing values and conceptions of harm. Thus, even for those who are ultimately unconvinced by the argument for a preference-eliciting—and thus narrow—interpretation of Section 230, the benefit of the preference-eliciting analysis is that it clarifies the stakes. Those who are unwilling to take the chance on the democratic process should be clear-eyed about the tradeoff they are willing to make: their conception of the benefits of the current judge-made immunity regime over the democratic process.

Conclusion

This paper has argued that the current, expansive interpretation of Section 230 is at best only one possible reading of the statute and, at worse, goes substantially beyond what Congress intended when it enacted the statute. The reason for this mismatch between the statute and its judicial interpretation is a result of repeated failures by Congress, the lower courts, and the Supreme Court, leading to a situation in which the courts, in trying to interpret this important statute, have no good options.

If courts allow their interpretation to be driven by their evaluation of the policy consequences, they will be unjustifiably substituting their own, potentially inaccurate, views over that of the democratic process. If they continue the status quo, they will lock in its benefits, but also its harms, and potentially lower the chance of congressional involvement. And if they interpret the statute narrowly, hoping to spur congressional action by mobilizing the political influence of giant internet platforms, they risk destabilizing the internet itself. Yet some degree of destabilization may be the necessary price to pay to put the law of intermediate liability on a firmer footing.

Alan Z. Rozenshtein is an Associate Professor of Law at the University of Minnesota (azr@umn.edu).