The Pipe, Not the Posts: How Section 230’s Protections Extend to Generative AI Platforms

Introduction

Section 230 of the Communications Decency Act (“Section 230”) protects online platforms from liability for user-generated speech. Based on the premise that platforms generally do not create original content but instead curate their users’ own speech, Section 230’s protections have historically applied to a platform’s algorithmically assembled, selected or sequenced content; algorithms or features that amplified or directed user generated speech to particular users were generally protected. Indeed, the old rule of thumb was the more a platform used algorithms or automation, the more likely user generated speech remained user generated speech. GenAI has the potential to change that calculus. A platform that ingests user generated content and publishes a transformed output, can blur Section 230’s line between publication and creation. The central question is whether Section 230 protects GenAI, which analyzes, but does not alter, user-created content to produce an outcome, and its corresponding outputs.

Section 230: The Basics

Congress enacted Section 230 in 1996 to shield online platforms from liability for user-created content. The “bedrock of the internet,” Section 230 allows for the open exchange of ideas and interactive content by protecting online platforms that deliver user content in a neutral way. Section 230(c)(2) also shields a platform’s good-faith efforts to moderate objectionable content.

While Section 230 protects platforms that publish content in a neutral way, exercising editorial judgment on how to deliver pre-existing content without evaluating its substance, it has not historically protected a platform that creates content by generating new or transforming user-generated content. The analogy to a pipe is often used. The owner of a pipe that merely transports water from point A to point B is not responsible for the content or quality of the water within. But,  what if the pipe owner decides who gets water, how much water, or influences the direction the water flows?  And what Section 230 has yet to tackle is what happens if the pipe owner adds its own water to the mixture, filters it, or manipulates the flow by using GenAI?   

Content-Neutrality

Several cases have held that Section 230 protects content-neutral functions. In other words, whether done by a human hand or by an algorithm, a platform’s decisions about others’ content, its placement in a feed, or whether it was pushed to particular users were all protected activities. For example, the Second Circuit concluded in Force v. Facebook, 934 F.3d 53, 66 (2d Cir. 2019), that Section 230 protects Facebook’s algorithms that suggested Hamas terrorist content to users. Algorithmic curation of existing user content is content-neutral. So long as Facebook did not create the content itself, then the promotion of or pushing out of the allegedly terrorist content or pushing it out to particular users was, in many ways, no different from a newspaper editor’s curatorial judgment as to where to place a story.

The Ninth Circuit in Dyroff v. Ultimate Software Group, Inc., 934 F.3d 1093, 1100-01 (9th Cir. 2019), reached a similar conclusion with respect to a social media platform’s features such as group recommendations and post notifications. These features were inherently content-neutral, and thus protected.

A few years later, yet before the current AI revolution, issues of algorithms and automation and Section 230 reached the Supreme Court in a duo of cases, Gonzalez v. Google, 598 U.S. 617 (2023) (per curiam) and Twitter, Inc. v. Taamneh, 598 U.S. 471 (2023). In the published majority Taamneh decision, the Supreme Court unanimously held that Twitter was not liable for its algorithm’s recommendation of ISIS content to users in connection with a terrorist attack. The Court found that the “passive” algorithm served as a delivery “pipe” rather than a “post,” operating at an arm’s length and “agnostic” to content, solely transmitting information made by users.

More recently, the Supreme Court addressed algorithmic platform curation in Moody v. NetChoice, LLC, 603 U.S. 707 (2024) (consolidated with NetChoice, LLC v. Paxton, No. 22-555). In Moody, the Supreme Court vacated a duo of lower court decisions that had upheld state laws that restricted platforms’ ability to “filter, prioritize, and label” third-party content. Id. at 717. Though the Supreme Court did not address Section 230 in its opinion, it nevertheless grappled with whether algorithmic content moderation reflects editorial judgment. As Justice Barrett flagged in her concurrence, AI used to identify and remove categories like “hateful” content raises questions about when an “algorithm” goes beyond implementing a human editorial choice. Id. at 746.

What Taamneh and Moody left open is what happens if the platform uses AI to do more than just filter, prioritize or label user-generated content, for example, actively recommends it or recharacterize its. In other words, is that adding water to the pipe, or merely deciding where the pipe should be placed?

Here Comes GenAI

GenAI presents an additional complication because it takes and analyzes existing content and creates a “new” output, but one that is inherently based on or derives from the original content. GenAI thus tests the literal water in the pipe that Taamneh and Moody left untouched. The Supreme Court in Taamneh validated the principle of content neutrality, but what is GenAI? Is it itself a form of content neutrality that merely pushes others’ content through the existing pipe, or is it itself content creation? 

To date, the lower courts have reached varying conclusions. At the District Court level, the Northern District of California has confronted the use of GenAI as a moderation tool. In Ryan v. X Corp., 2024 WL 5058526 (N.D. Cal. Dec. 9, 2024), the court found that using GenAI to moderate content did not strip a website of its Section 230 protections. It went as far as to say that no authority supported the position that Section 230 did not protect interactive platforms using AI. And it found that the majority in Moody did not address the use of GenAI to regulate content.

On the other side of the ledger, the same court has found that using GenAI to enhance or revise content, particularly advertising, on a platform, might be outside the scope of Section 230. In both Forrest v. Meta Platforms Inc., 737 F. Supp. 3d 808 (N.D. Cal. 2024) and Bouck v. Meta Platforms, Inc., 2026 WL 810036 (N.D. Cal. Mar. 24, 2026), the court distinguished between merely publishing material and contributing sufficiently to the content to render Section 230 inapplicable. But the court did not offer a fixed rule in either case, finding that whether the contributions were sufficient for liability was a question of fact.

The Third Circuit has taken a different tack, holding in Anderson v. TikTok, Inc., 116 F.4th 180 (3d Cir. 2024), that TikTok’s “For You Page” recommendation algorithm, which showed a “Blackout Challenge” video to a minor, and used the minor’s metadata, demographics, and online interactions, fell outside Section 230’s immunity. If recommendation engines lack Section 230 protection, other GenAI features may be at risk too.

What Platforms Should Do Now

Whether Section 230 is a defense to claims based on a platform’s features that rely on or use outputs from recently released generative AI products that consciously evaluate, transform, or summarize content is an issue of first impression that courts will tackle soon. Examples abound:

• Platforms evaluating user reviews to issue a summary may be considered new content.
• Chatbots synthesizing user data to provide advice or a conclusion may not be protected.
• Recommendation engines based on user generated speech, demographics, or metadata may be considered the platform’s speech.

As the Section 230 analysis is highly fact-specific, platforms that want to preserve the strongest possible immunity argument should build and document their GenAI features, keeping in mind:

1. Content-Neutrality. Section 230 turns on the neutrality of a tool. For GenAI, that means avoiding prompt design, rewrites, and defaults steering outputs based on their content.
2. Hosting vs. Generating. The closer GenAI is to organizing user-generated content, rather than creating a new statement, the stronger the publisher defense under Section 230.
3. Label and preserve. Label AI vs. user-generated outputs, and document guardrails against hallucinations to ensure any GenAI output can be shown to be derivative of user speech.

Platforms should ensure that GenAI algorithms are grounded in intentional design, and deployed to avoid generating new or edited content based on user speech. The more the platform is endorsing, recommending, or directing specific traffic or users to a particular post or speech, as opposed to merely publishing it for large groups to see or surfacing it in their feed based on their own engagement with the platform, the more likely those design choices may risk scrutiny and invoke a Section 230 challenge. Documenting design choices and user-facing representations is therefore necessary to ensure that GenAI remains a pipe for delivering content, rather than a post that alters content before it reaches users’ screens.

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.