Beyond the Publisher: Rethinking Section 230 Immunity for Algorithmic Conduct

Introduction

Since the rise of Facebook (Meta), X, TikTok and YouTube, social media platforms have become the primary means through which billions of people communicate and consume information. As these platforms increasingly influence public discourse through recommendation algorithms and content moderation, courts and legislators have questioned whether they should bear the same legal responsibilities as traditional publishers.

Section 230 of the Communications Decency Act (1996) is a foundational U.S. law that shields online platforms and users from legal liability for third-party content.[1] It essentially prevents online service providers from being classified as publishers, allowing them to moderate content in good faith.

But the rise of advanced recommendation algorithms reveals a limit in the traditional idea of intermediary immunity. Although Section 230 shields platforms from liability for third-party content, it’s unclear whether that protection extends to harms caused by the platform’s own algorithms.[2] Therefore, this article contends that while Section 230 rightly shields platforms from liability for user-generated content, it should not cover a platform’s own algorithmic choices that significantly lead to foreseeable harm.[3]

The article first considers the legal foundations of intermediary immunity before examining recent case law on algorithmic recommendations. It concludes by arguing that platform liability should be confined to harms arising from a platform’s own technological conduct, rather than the user-generated content it hosts.

Section 230’s Foundations and Its Limits

Section 230(c)(1) states that a ‘provider or user of an interactive computer service will not be considered a publisher or speaker of content [of a third party]’,  meaning platforms like Facebook, Instagram, and YouTube cannot be sued for defamatory, fraudulent, or illegal posts made by users.[4] Additionally, under (c)(2), companies can voluntarily remove, hide, or moderate ‘objectionable’ user-generated content without risking legal liability for these moderation decisions.[5] This law was initially enacted to overturn the ruling in Stratton Oakmont, Inc. v. Prodigy Services Co., in which the court emphasised that a ‘conscious choice’ to exercise editorial control, such as implementing policies, could make an internet service provider liable as a publisher.[6] As a result, until 1995, digital service providers such as Prodigy were held liable. However, even then, sponsors of the bill, such as Representative Christopher Cox, argued that this decision was ‘backward’ and encouraged providers to actively manage the content displayed to the public.[7]

Thousands of digital platforms face lawsuits. Rather than suing for the user content itself (which is shielded by legal immunities), the recent wave of litigation targets the algorithmic design that promotes harmful content or makes platforms intentionally addictive. The core tension is that the statute is thirty years old and written for message boards and forums, not algorithmic recommendation engines. Courts split largely on whether a platform’s own algorithm counts as third-party content (protected) or as the platform’s own ‘expressive product’ (not protected). Max Del Real in Article 1 of the Washington Law Review (Vol 99)  catoagorises algorthimic output as either ‘(1) standalone information… or (2) as augmentative to the recommended third-party information,’ and argues that if the algorithmic message causes or is the basis for an illegal claim, section 230 immunity does not apply because the defendant isn’t considered the publisher of third-party content.[8]

III. A Circuit Split: Algorithms as Third-Party Content or the Platform’s Own Speech

Despite varying scholarly opinions and past interpretations of Section 230, some cases conflict, leaving the fundamental legal question about whether algorithmic recommendations are excluded from Section 230 unresolved. [9]

The Third Circuit ruled in Anderson v. TikTok that Section 230 doesn’t shield platforms from harm caused by their own algorithmic recommendation of third-party content, reasoning that under the Supreme Court’s Moody v. NetChoice decision, a platform’s curatorial choices are its own protected expression and if it’s the platform’s own speech for First Amendment purposes, it’s the platform’s own speech for Section 230 purposes too, meaning immunity doesn’t apply.[10]

The Fourth Circuit reached the opposite conclusion in M.P. v Meta Platforms, treating algorithmic curation as simply a tool for organising and surfacing third-party content, and therefore still within the scope of Section 230 immunity regardless of how personalised the feed is.[11] A Berkeley Law review of developments in 2025–2026 identified these two decisions as directly irreconcilable, describing a live Third/Fourth Circuit split on algorithmic-feed product liability.[12]

In Force v. Facebook, the Second Circuit majority held that an algorithm directing users toward extremist content was still protected by Section 230.[13] However, Chief Judge Katzmann questioned the majority’s treatment of Facebook’s algorithm.[14] Katzmann argues that ‘Facebook uses algorithms to create and communicate its own message: that it thinks you, the reader—specifically—will like this content,’ contending that this extends Facebook’s role beyond the traditional publishing defined in Section 230, and doing so is ‘directly related to the alleged illegality of the site’.[15] Del Real notes Katzmann’s view as the ‘first judicial acknowledgement’ of algorithmic recommendations not being protected by section 230 beacsue of the way companys ‘use them to communicate their own Messages’.[16]

The Supreme Court had a chance to resolve this in Gonzalez v. Google, but it punted, issuing a narrow per curiam decision that resolved the case on other grounds without ever answering whether algorithmic recommendations fall outside Section 230, leaving the underlying legal question open.[17]

The Scholarly Divide: For and Against Narrowing Immunity

Del Real’s article ultimately argues that the algorithm’s output is not just third-party content hosted on a platform but a unique product created by the platform’s own processing. He describes it as either ‘standalone information’ produced by the platform or as the platform’s own material added onto third-party content.[18] Therefore, digital platforms are not just hosting content; their recommendation systems are an independent product of the platform and should not be protected under publisher immunity.

Not all scholars agree that narrowing Section 230 for algorithmic conduct is desirable. Jennifer Huddleston wrote in the George Mason Law Review, arguing that the Third Circuit’s reasoning in Anderson v TikTok risks creating a ‘new moderator’s dilemma,’ warning that denying immunity to algorithmic curation ‘significantly narrows the existing interpretation of Section 230 and renders it inversely related to the First Amendment.’[19] On her account, removing algorithmic protection would push platforms toward two equally undesirable extremes: costly human-only moderation that cannot scale, or abandoning moderation tools altogether, leaving users exposed to more ‘lawful but awful’ content, not less.[20] She also highlights a competition concern that reduced immunity could disproportionately impact smaller platforms, since only the largest companies can withstand ongoing legal expenses, thereby strengthening existing incumbents rather than challenging them.[21]

However, Huddleston’s warning targets a blanket removal of algorithmic immunity. It does not address a narrower standard, especially one that preserves protection for ordinary curation but withdraws it only where a platform’s own design choices foreseeably and materially contribute to harm, which is the distinction this Article defends.

As Danielle Keats Citron in the Boston University Law Review reminds us, Section 230’s animating purpose was never unconditional protection: ‘the purpose of section 230 was to incentivise responsible content moderation, not to shield from liability sites that make a mockery of the concept’.[22] The same logic extends naturally to platforms whose algorithmic design foreseeably amplifies harm; such conduct is no more consistent with section 230’s original purpose than the deliberate bad-faith hosting Citron describes.

Conclusion

In the bigger picture, underneath the doctrine, courts disagree because they’re balancing two different things Congress arguably wanted at once: protecting free expression online by not treating platforms like publishers, and giving injured plaintiffs (often families of harmed minors) some avenue for relief when a platform’s own design choices (not just user content) caused harm. As algorithms have become more central to how people encounter content, that balance is genuinely contested, and the emerging circuit split may prompt the Supreme Court to take up the question again, hopefully resolving it this time rather than punting as it did in Gonzalez.[23]

Bibliography: 

Primary sources: 

Cases: 

– Anderson v TikTok Inc, 116 F4th 180 (3d Cir 2024)

– Force v Facebook, Inc, 934 F3d 53 (2d Cir 2019)

– Gonzalez v Google LLC, 598 US 617 (2023)

– MP v Meta Platforms Inc, 127 F4th 762 (4th Cir 2025)

– Moody v NetChoice LLC, 144 S Ct 2383 (2024)

– Stratton Oakmont, Inc v Prodigy Services Co, No 31063/94, 1995 WL 323710 (NY Sup Ct, 24 May 1995)

Statutes: 

– Communications Decency Act 1996, 47 USC § 230

Secondary sources: 

Journals:

– Citron DK, ‘How To Fix Section 230’ (2023) 103 BU L Rev 713

– Del Real M, ‘Breaking Algorithmic Immunity: Why Section 230 Immunity May Not Extend to Recommendation Algorithms’ (2024) 99 Wash L Rev Online 1

– Huddleston J, ‘Anderson, Algorithms, and Section 230 After NetChoice: The Risk of a New Moderator’s Dilemma’ (2026) 32 Geo Mason L Rev 727

Official Publications (websites): 

– UC Berkeley Law, ‘Internet and Computer Law Year in Review 2025–2026: AI Copyright Fair Use, the CDA Circuit Split, and the Age Verification Flood Gate’ (Berkeley Centre for Law & Technology), < https://www.law.berkeley.edu/research/bclt/bclt-legal-analysis/internet-computer-law-s1/>, accessed 9 July

Congressional Materials:

– 141 CONG. REC. H8470 (1995) (statement of Rep. Christopher Cox) < https://www.congress.gov/crs-product/R46751#_Toc155275795 > accessed 9 July

[1]  Communications Decency Act 1996, 47 USC § 230.

[2] Ibid.

[3] Ibid.

[4] ibid § 230(c)(1).

[5] ibid § 230(c)(2).

[6] Stratton Oakmont, Inc v Prodigy Services Co, No 31063/94, 1995 WL 323710 (NY Sup Ct, 24 May 1995).

[7] 141 CONG. REC. H8470 (1995) (statement of Rep. Christopher Cox).

[8] Max Del Real, ‘Breaking Algorithmic Immunity: Why Section 230 Immunity May Not Extend to Recommendation Algorithms’ (2024) 99 Wash L Rev Online 1, 31.

[9]  CDA (n1).

[10] Anderson v TikTok Inc, 116 F4th 180 (3d Cir 2024); Moody v NetChoice LLC, 144 S Ct 2383 (2024).

[11] MP v Meta Platforms Inc, 127 F4th 762 (4th Cir 2025).

[12] UC Berkeley Law, ‘Internet and Computer Law Year in Review 2025–2026: AI Copyright Fair Use, the CDA Circuit Split, and the Age Verification Flood Gate’ (Berkeley Center for Law & Technology), < https://www.law.berkeley.edu/research/bclt/bclt-legal-analysis/internet-computer-law-s1/>, accessed 9 July 2026.

[13] Force v Facebook, Inc, 934 F3d 53 (2d Cir 2019).

[14] Ibid, 76 (Katzmann, J., concurring in part).

[15] Ibid, 82-83.

[16] Del Real (n8)  23.

[17]  Gonzalez v Google LLC, 598 US 617 (2023) ; Del Real (n8) 30.

[18] Del Real (n8) 30-36.

[19] Jennifer Huddleston, ‘Anderson, Algorithms, and Section 230 After NetChoice: The Risk of a New Moderator’s Dilemma’ (2026) 32 Geo Mason L Rev 727, 734 (Part II).

[20] Ibid, 735.

[21] Ibid, 737-738.

[22] Danielle Keats Citron, ‘How To Fix Section 230’ (2023) 103 BU L Rev 713, 750.

[23] Gonzalez (n17).