Letter from 42 attorneys general on sycophantic and delusion

Letter to AI companies on sycophantic and delusional outputs

Letter. Press release headline (State attorneys general (42)): “AG Sunday Leads Coalition of 42 Attorneys General in Letter to A.I. Software Companies Demanding Safeguards to Protect Vulnerable Residents from Harmful Interactions with Bots”

Enforcement action

Allegations. This is a letter from state attorneys general. It states their concerns and allegations; it is not a lawsuit or a finding that anyone broke the law. Status as of October 21, 2026.

Letter dated Dec 9, 2025; announced Dec 10, 2025. Companies were asked to commit to the requested changes by Jan 16, 2026 and to meet with the attorneys general.

Chatbot Laws · verified Oct. 21, 2026

Forty-two attorneys general, led by Pennsylvania with New Jersey, West Virginia and Massachusetts, wrote to 13 AI companies about sycophantic and delusional chatbot outputs, which the letter links to deaths and hospitalizations, and about harmful chatbot conversations with children. They asked for safety testing, recall procedures, clear warnings and stronger child protections, with a commitment by Jan 16, 2026.

Addressed to

Anthropic, Apple, Chai AI, Character Technologies, Google, Luka, Meta, Microsoft, Nomi AI, OpenAI, Perplexity AI, Replika and xAI.

What the action covers

Other duties	Warning letter, not a lawsuit. It asks companies to adopt policies against sycophantic and delusional outputs, safety-test models before release, recall models whose dangerous outputs cannot be stopped, show permanent warnings on the input screen, separate revenue goals from safety decisions, allow independent audits, notify users exposed to harmful outputs, publish incident logs and safety test results, block outputs to child accounts that encourage grooming, drug use, violence, self-harm or secrecy from parents, and publish a protocol for reporting such risks.Letter § III

Other duties

Warning letter, not a lawsuit. It asks companies to adopt policies against sycophantic and delusional outputs, safety-test models before release, recall models whose dangerous outputs cannot be stopped, show permanent warnings on the input screen, separate revenue goals from safety decisions, allow independent audits, notify users exposed to harmful outputs, publish incident logs and safety test results, block outputs to child accounts that encourage grooming, drug use, violence, self-harm or secrecy from parents, and publish a protocol for reporting such risks.Letter § III

Who brought it and what it seeks

Sent by	State attorneys general (42)
Penalties	None; warning letter. The letter says failing to add safeguards may violate state laws on warnings, defective products, unfair or deceptive practices and children's online privacy.

Related laws

Letter from 44 attorneys general to AI companiesLetter from state attorneys general
Announced Aug. 25, 2025

Sources

Pennsylvania Attorney General release, Dec 10, 2025 Supports: Lead and co-leads, count of AGs, recipients, demands, Jan 16, 2026 deadline.
West Virginia Attorney General release, Dec 10, 2025 Supports: Co-leads, recipients, demands, deadline.
Connecticut Attorney General release, Dec 12, 2025 Supports: Co-leads, count of AGs, recipients.
New York Attorney General release, Dec 10, 2025 Supports: Count of AGs; link to the letter.
The letter, with signatures (hosted by the New York AG) Supports: Date, recipients, signers, concerns, the 16 requested safeguards, deadline.
The letter (hosted by the Iowa AG) Supports: Date, recipients, signer count.