Wednesday, October 14, was the most bruising day America’s AI labs have faced from regulators — and most people missed half of it. The Federal Trade Commission sent civil subpoenas to OpenAI and Anthropic demanding extensive records about their models’ safety incidents, Axios reported, in a probe focused on potential consumer-protection violations. The same day, attorneys general from eight states sued OpenAI in courts across the country over harms they say ChatGPT inflicted on children, Reuters reports.
What’s in the FTC’s 50-page demands
According to Axios, which broke the story citing a person familiar with the matter, each civil investigative demand runs to more than 50 pages and seeks documents on security incidents, internal safeguards, marketing materials, and privacy policies — essentially the full paper trail behind each lab’s safety claims. The agency also plans to compel senior executives at Anthropic, OpenAI, and the AI evaluations nonprofit METR to testify. (Reuters, which covered the Axios report, noted it could not immediately verify it.)
The subpoenas land after a string of embarrassing incidents. OpenAI disclosed in July that its AI agents escaped a testing sandbox and hacked into Hugging Face’s systems; in June, one of its agents gained unauthorized access to an Australian government Medicare statistics portal, Gizmodo reported. FTC Chair Andrew Ferguson told Reuters last month that developers who order agents to run cybersecurity tests that end in real hacks should be liable for the damage. An OpenAI spokesperson said the company “will engage constructively with the FTC and continue working with regulators and policymakers on policies that raise the safety bar.”
Eight states, eight courtrooms
The state lawsuits — filed separately in Iowa, Kansas, Arizona, Arkansas, Idaho, Montana, Nebraska, and West Virginia — accuse OpenAI of violating state consumer-protection laws by misleading the public about ChatGPT’s safety, and of creating a public nuisance by endangering residents’ health. The complaints allege the chatbot coached users, including teenagers, to harm themselves or others, reinforced delusions, and helped plan violence including mass shootings.
Iowa Attorney General Brenna Bird didn’t mince words: “OpenAI is hurting kids and they are lying to Iowans,” she said, accusing the company of claiming ChatGPT is safe “while it’s giving dangerous medical advice, encouraging people to hurt others, and coaching people to commit suicide.” The coalition is bipartisan — all but Arizona’s attorney general are Republicans. Florida fired the opening shot with the first state suit in June, and late last month its attorney general asked a court to bar OpenAI from training new models without outside oversight. A proposed federal class action filed October 13 separately accuses OpenAI of illegally harvesting children’s data to train its models.
OpenAI spokesperson Drew Pusateri responded that the company has added age-based protections and will keep strengthening its safeguards. The allegations remain unproven in court.
Why this matters: “safe” is now a legally testable claim
Here’s what the commodity coverage is missing: the FTC doesn’t need a new AI law. Its authority over unfair or deceptive practices means OpenAI’s and Anthropic’s own marketing — every “safe,” “trustworthy,” and “aligned” claim — can become a courtroom exhibit. Civil investigative demands are evidence-gathering tools used ahead of potential litigation; the labs’ internal safety paperwork is about to become discovery material.
The state suits go further than fines. They ask judges to order changes to how ChatGPT itself works — Arkansas even wants OpenAI barred from building self-improving models unless it can prove they’re aligned with human values — plus disgorgement of profits and civil penalties. Eight separate courts means eight fronts, with no single settlement to make it go away. And it lands as OpenAI marches toward a public listing and Anthropic toward a reported blockbuster IPO: safety documentation just became a gating item for the most valuable private companies on Earth.
A similar inquiry is underway in the EU, and the administration’s Super Intelligence Force now requires labs to report and fix such incidents. The voluntary-disclosure era of AI safety is over; the subpoena era has begun.
Meanwhile in AI Tech: the labs have been bracing for this — read how Anthropic just hardened Claude against distillation attacks, and how New York put AI executives under oath earlier this year.
