Section 230 of the Communications Decency Act, codified at 47 U.S.C. § 230, is the United States intermediary-liability statute providing that an interactive computer service may not be treated as the publisher or speaker of content supplied by another information content provider. Whether that protection reaches output generated by an AI system — rather than hosted from a third party — is unresolved and is being litigated across several AI-harm cases.
Statutory structure
The operative provisions are two subsections. § 230(c)(1) provides that "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." § 230(c)(2) supplies good-faith content-moderation immunity, the clause identified as the "Good Samaritan" provision (An Introduction to Section 230 — Eric Goldman (Section 230 Primer, 2022)).
Before Section 230, liability for third-party content attached where the disseminator exercised editorial discretion: common carriers such as telephone providers were not liable, while traditional publishers were. Online services broke that binary by both carrying and editing content (An Introduction to Section 230 — Eric Goldman (Section 230 Primer, 2022)). Section 230's shield enabled the open-platform architecture of the commercial internet, and its scope is a precondition for most US platform-regulation debate (An Introduction to Section 230 — Eric Goldman (Section 230 Primer, 2022)).
The third-party-content predicate and AI output
The statute conditions immunity on the content originating with "another information content provider." That predicate is the fault line for AI systems, because a generative model produces the text at issue rather than hosting it.
Commentators have made this argument directly. Following the Munich Regional Court's ruling against Google over false statements in AI Overviews, Gary Marcus argued that Section 230-style immunity may not shield AI-generated content in the United States, since a generated summary is the platform's own speech rather than hosted third-party speech (Source: garymarcus.substack.com; see LG München I — Google AI Overviews liability (26 O 869/26)). Legal observers covering Florida's suit against OpenAI made a parallel point: the suit's product-liability theory tracks the litigation wave against social-media platforms, but Section 230 may afford AI companies weaker protection because the chatbot itself produces the speech (Source: politico.com; see Florida v. OpenAI).
The question has been framed in the Character.AI Litigation (umbrella) as whether Character.AI qualifies as a publisher of third-party content under § 230, or whether the AI character is the speaker and Character.AI the speaker's principal. That framing remains the operative but unresolved question; the January 2026 settlements removed the vehicle that would have carried it to appellate treatment in Garcia, so the anticipated appellate ruling on the Section 230 and product-liability questions did not arrive.
In Raine v. OpenAI, Inc., the suit advanced past the motion-to-dismiss stage on April 22, 2026, defeating OpenAI's threshold defenses on Section 230 and product-liability framing, and proceeded to discovery (Source: courthousenews.com). In Garcia v. Character Technologies, Inc., the May 2025 ruling rejected categorical Section 230 protection for chatbot outputs (An Introduction to Section 230 — Eric Goldman (Section 230 Primer, 2022)). Section 230 is also implicated in agent-platform liability questions raised by Amazon v. Perplexity AI, alongside the surrounding computer-trespass doctrine (Norms of Computer Trespass — Orin S. Kerr (Essay, 2022)).
The features-versus-content line
A second line of dispute concerns whether a claim targets content or product design. In New Mexico v. Meta Platforms (youth harm and chatbot decree), Judge Biedscheid rejected Meta's Section 230 defence, finding that the state challenged platform features rather than third-party content; the same judge then invoked Section 230 among his grounds for declining to order changes to Meta's algorithms, infinite scroll and autoplay (Source: straitstimes.com). The line was therefore drawn within the case rather than around it. This features-versus-content distinction is the same one the design-defect theory runs on across the youth-harm cluster (Content vs Architecture Theory of Social Media Harm).
Interaction with statutory duties
Recent legislation has tended to layer duties alongside Section 230 rather than amend it. The TAKE IT DOWN Act does not amend Section 230 directly; it operates through a statutory duty that platforms must meet independent of Section 230 immunities, a structure differing from the DMCA and from FOSTA-SESTA. Utah's Utah HB 276 — Digital Voyeurism Prevention Act + Digital Content Provenance Standards Act states that it does not alter, modify, or limit Section 230, meaning the state cannot impose liability on a covered platform for acting as the publisher or speaker of third-party content. Commentary has placed these alongside FOSTA-SESTA and the Kids Online Safety Act (KOSA, S. 1748) as a multi-year narrowing of intermediary protection.
Effect on moderation incentives
Because Section 230 largely shields platforms from liability for user content, legal pressure to moderate accurately is reduced, and platforms tend to moderate for brand safety rather than legal compliance; the argument that AI training choices are protected speech adds a further layer of insulation from regulation (AI and Content Moderation, AI and the First Amendment).
Open questions
- Whether the Section 230 and product-liability questions in the companion-chatbot cases reach an appellate ruling is undetermined following the January 2026 settlements (Character.AI Litigation (umbrella)).
- Whether courts will treat model output as the service's own speech, as third-party content restated, or as a category the statute does not address has not been settled by any appellate decision reported to date.
Relationships
- depends-on: An Introduction to Section 230 — Eric Goldman (Section 230 Primer, 2022) — doctrinal reference for the statutory structure and pre-1996 baseline
- related: AI and Content Moderation, AI and the First Amendment, Algorithmic Speech Doctrine, AI Liability, Content vs Architecture Theory of Social Media Harm, Norms of Computer Trespass — Orin S. Kerr (Essay, 2022)
- instance-of: AI Liability — intermediary-liability regime within the broader AI liability question
- contested-in: Garcia v. Character Technologies, Inc., Character.AI Litigation (umbrella), Raine v. OpenAI, Inc., Florida v. OpenAI, New Mexico v. Meta Platforms (youth harm and chatbot decree), Amazon v. Perplexity AI
- related: TAKE IT DOWN Act, Utah HB 276 — Digital Voyeurism Prevention Act + Digital Content Provenance Standards Act, Kids Online Safety Act (KOSA, S. 1748), Eric Goldman, Orin S. Kerr