Question 1

What did Walters v. OpenAI actually decide about LLM defamation liability?

Accepted Answer

Walters v. OpenAI was dismissed at summary judgment by Gwinnett County Superior Court Judge Tracie Cason in May 2024 on three separate grounds that together establish a high but not impossible bar for plaintiffs. First, the court held that a reasonable reader would not understand a ChatGPT output to be a statement of fact given OpenAI's disclaimers about hallucination risk in the product interface. Second, the court found Walters could not show actual malice because he was a public figure on radio and OpenAI had no knowledge of falsity or reckless disregard. Third, the court found no actual damages because the only recipient of the false output was the journalist who recognized the error and never published it. The case did not resolve whether LLM outputs can ever be defamatory; it resolved that this particular output to this particular plaintiff was not. Future plaintiffs with private-figure status, downstream publication, and provable damages remain a live risk.

Question 2

Can a brand sue an LLM provider for false information about its products or executives?

Accepted Answer

Yes, in principle, and several active cases test the theory in 2026. The viable claims fall into three buckets: trade libel for false statements about products that cause measurable revenue loss, false advertising under Lanham Act Section 43(a) for AI outputs that misrepresent a competitor or the plaintiff's own brand in ways tied to commercial transactions, and traditional defamation for false statements about identifiable executives that injure professional reputation. Trade libel and Lanham Act claims have lower First Amendment friction than personal defamation because they implicate commercial speech rather than reportage. The hard part for brand plaintiffs is causation: showing that a specific false LLM output caused a specific lost deal or measurable trust damage. Brands that win these cases will be the ones who instrumented citation monitoring early and have evidence linking specific outputs to specific buyer decisions.

Question 3

How is NYT v. OpenAI different from defamation cases, and why does it matter for liability?

Accepted Answer

NYT v. OpenAI, filed December 2023 in the Southern District of New York, is a copyright and trademark case, not a defamation case, but it matters for liability because the discovery and damages framework being built there will be borrowed by every plaintiff with an AI-output complaint. The case turns on whether training on copyrighted articles without license constitutes infringement, whether ChatGPT's verbatim regurgitation of paywalled NYT content is fair use, and whether OpenAI's attribution failures constitute Lanham Act false designation of origin. Judge Sidney Stein has allowed the core claims to proceed past motion to dismiss in March 2025, signaling that volume-of-training and downstream-output theories will both get full discovery. The trademark dilution and false designation theories from NYT v. OpenAI are the same theories brand defamation plaintiffs will rely on in 2026 and 2027.

Question 4

What should a company do if ChatGPT or Claude publishes false information about its brand?

Accepted Answer

Move on three tracks simultaneously and document every step. Track one is a formal correction request through the model provider's content reporting channel (OpenAI Trust and Safety, Anthropic abuse reporting, Google Trust and Safety) with screenshots, the exact prompt, the date and time, and the requested remediation. Track two is corrective publication: a clear, schema-tagged, dated page on the company website that authoritatively states the correct fact, optimized to be the highest-confidence source the next time the model retrieves the topic. Track three is preservation of evidence including timestamped screenshots, archived prompts and responses, and contemporaneous notes on any customer or partner exposure. The preservation track is the one most operators skip and the one that determines whether a defamation claim is viable 18 months later when damages have accumulated and counsel is needed.

Question 5

Will Section 230 protect LLM providers from defamation claims for generated content?

Accepted Answer

Almost certainly not, based on the Supreme Court's reasoning in Moody v. NetChoice (July 2024) and the Third Circuit's holding in Anderson v. TikTok (August 2024) that algorithmic curation choices can constitute first-party speech rather than third-party publisher conduct. The traditional Section 230 immunity applies when a platform passively hosts user content without material contribution. LLM providers actively generate outputs through model weights they trained and tuned, which courts are increasingly treating as a form of authorship rather than hosting. Walters v. OpenAI explicitly declined to rely on Section 230 even though OpenAI raised the defense. The defamation defense bar in 2026 has largely shifted to disclaimer-and-context arguments under Milkovich v. Lorain Journal rather than statutory immunity, which is a structurally weaker position because it requires fact-specific analysis of each output.