The Pipe, Not the Posts: Part Deux

In our previous client alert, we addressed how algorithms and platform features that rely on user generated speech might not receive Section 230 protections because those features may not be considered platform user content.  In these instances, the platform might not carry user generated water through the pipe, but adds its own to the mixture.  At the same time, if it is true that there is less Section 230 protection, these same elements should then imbue the content with First Amendment protections as the expressive content of the platform.

A recent decision calls into question whether Courts will see the trade off in this way.  Judge Edward J. Davila’s August 5, 2026 order denying motions for preliminary injunction to block California from enforcing the Protecting Our Kids from Social Media Addiction Act (SB 976), which restricts social media platforms from providing addictive, algorithm-driven feeds and timed notifications to minors without verifiable parental consent, in Meta Platforms, Inc., et al v. Rob Bonta, No. 5:25-cv-09792-EJD (N.D. Cal. Aug. 5, 2026), calls this assumption into question, and has potential implications for Section 230 immunity and First Amendment protection for online algorithms.

I. The First Amendment, Section 230, and Moody

Section 230 provides immunity to platforms that deliver user created content.  Courts are increasingly grappling with whether algorithms and AI curated content are subject to Section 230 protections, and concluding that they are not.  See State ex rel. Torrez v. Meta Platforms, Inc., No. D-101-CV-2023-02838 (N.M. 1st Jud. Dist. Ct. Aug. 6, 2026) (finding Section 230 does not protect platform product design, algorithms and engagement features).

But if not protected by Section 230, the question becomes whether those things the platform uses to curate content are protected by the First Amendment.  The theory is that the more the content is that of the platforms (and not subject to Section 230), the more the content becomes the speech of the platform subject to First Amendment protections. 

In the era of AI and algorithms, this distinction is harder to draw, as algorithms that platforms use to curate or publish user speech may be bespoke and esoteric products in and of themselves that rely on intricate design, code, and other weights and measures.  The Supreme Court acknowledged this in Moody v. NetChoice, LLC, 144 S. Ct. 2383 (2024) through a concurrence noting that algorithms can be original expressions worthy of First Amendment protection.  Id. at 2410 (Barrett, J., concurring).  The recent decision in Meta v. Bonta seems to confound this issue, rather than bringing clarity.

II. Meta v. Bonta

Plaintiffs Meta, TikTok, Google, and YouTube challenged the Protecting Our Kids from Social Media Addiction Act (SB 976) claiming that it sought to control content created by the platforms’ algorithms, which suggested videos to users based on user history.  The platforms argued that their curation of content, and therefore their First Amendment rights, were at issue.

Judge Davila rejected this argument, concluding that the algorithms did not promote the speech of the platforms, but promoted the speech of the users.  The Court characterized the feed general process “as [a] ‘passive receptacle[s]’ of third-party speech or as ‘dumb pipes’ that merely emit what they are fed.”  Meta Platforms, Inc., et al v. Rob Bonta, No. 5:25-cv-09792-EJD at *10 (N.D. Cal. Aug. 5, 2026) (citing Moody v. NetChoice, LLC, 144 S. Ct.  (2024) (Alito, J., concurring)).  Judge Davila found that because the algorithms suggest content based on a user’s own engagements and preferences, it is user speech, analogizing the feeds to “a mirror that reflects back to users their own perceived interests,” rather than reflecting moral judgments embodied in Plaintiffs’ Community Guidelines.  Id. at *8. 

Judge Davila did not focus on the many steps, factors, and processes that comprise an algorithm, and instead focused on the algorithm’s (1) chief input—the user generated content itself; and (2) its chief purpose—surfacing more of that user generated content.  And thus he concluded that the algorithm—the process at work—does not give the right to the platform to assert a First Amendment claim.  Judge Davila’s decision does not address Section 230, yet his holding has implications for its effect on platform products such as algorithms.

III. What Does This Mean for Section 230? 

If the algorithm is not protected speech because it is a mirror or output of user generated speech, it should garner Section 230 protection.  But if not user content, then algorithms and curated content would appear to be subject to First Amendment protection.  See, e.g., Miami Herald Publishing Co. v. Tornillo, 94 S. Ct. 2831 (1974).

While the issue of curation is more or less settled law in the pre-AI (analog) world, it is less so when AI curation and content creation by algorithms is at issue.  Compare Moody v. NetChoice, LLC, 144 S. Ct. 2383, 2410 (2024) (Barrett, J., concurring) (noting that algorithms may receive First Amendment protections) with Moody v. NetChoice, LLC, 144 S. Ct. 2383, 2410 (2024) (Alito, J., dissenting) (AI curated moderation is not seen as protected by First Amendment). 

Judge Davila’s decision underscores that AI products themselves can be a mixture of curation and creation.  Indeed, what if certain algorithms, especially those with GenAI outputs, are determined by courts to be partially curation and partially creation?  Are platform features that take user generated content and add to it, modify it, or generate new outputs at risk of losing both Section 230 immunity and First Amendment protection?

IV. Takeaway

The ruling in Meta v. Bonta is on appeal, and Judge Davila has since denied the platforms’ motion to stay enforcement, reaffirming that their algorithms are not protected expression.  This reinforces the importance of separating the publishing and curation of third-party content from a platform’s own content creation.  The flipside to Bonta is that while the algorithm that merely surfaces or mirrors user generated content is unlikely to qualify as First Amendment protected speech (at least under the analysis adopted in Meta), it should garner Section 230 immunity.  Although Meta v. Bonta does not address Section 230, it stands to reason that platform features that take user generated content and add to it, or create new outputs may risk losing both Section 230 and First Amendment protections.

Companies integrating GenAI should ensure that features generating new content are clearly distinguished from passive recommendation engines, both in design and documentation, as the further a GenAI feature moves toward producing original output, the more it risks losing both First Amendment protection as platform speech and Section 230 immunity as a neutral publisher.

This article is intended to provide general information about legal developments and does not constitute legal advice. Readers seeking guidance with respect to specific legal matters should consult with counsel.