FTC investigates OpenAI and Anthropic a day after Trump's AI accord asked labs to police themselves

The Federal Trade Commission is investigating OpenAI, Anthropic and other AI companies over the safety of their products, an agency spokesperson told Axios on 30 September. The confirmation came a day after President Trump and the heads of six AI companies signed a one-page White House accord promising to police themselves, a pledge Trump called "morally binding" and which carries no penalties. The FTC's probe opened this summer, and its chair is preparing demands that could compel executives to hand over documents and testify.

What is the White House Accord on Super Intelligence?

It is a voluntary, one-page pledge by six AI companies to run four layers of safety controls on their own models. Its full title is the White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities, signed after a White House lunch on 29 September. It opens: "we believe every company is responsible for developing its own technology safely and in a way that builds trust with customers and the public."

The name follows the administration's own relabelling. In the US-China fact sheet four days earlier, both governments agreed to call the technology "super intelligence" rather than "artificial intelligence", as we reported in our hotline item.

What do AI companies have to do under the White House accord?

Four things, all inside the company. The accord says each company "should implement the following four layers of controls and audits":

1. Internal controls to monitor models "during training and deployment around areas like cybersecurity, biosecurity, and chemical threats, and to ensure that its models do not hack or access technical systems in unintended ways."
2. An internal team to make sure the controls, monitoring and detection work, "and that any issues are remediated."
3. "An independent external auditor or evaluator to carry out independent assessments" of whether the controls work.
4. "An independent committee of the board of directors" to receive reports from those teams and auditors and see that issues are fixed.

The companies also agree to "meet regularly to establish standards and best practices." Axios writes that the rules are similar to what the companies were already doing.

Is the White House AI accord legally binding?

No. Al Jazeera reports that the pact is voluntary and not legally binding, sets no penalties for companies that fail to comply, does not require them to publish audit results, and gives the government no enforcement role. "There's going to be a tremendous self-policing aspect," Trump told reporters, adding that it is "morally binding."

The text leaves a door open: "Over time, it may make sense to codify these steps into laws or regulations." It does not say who would do that or when. Forbes noted that "United States" is misspelled "Unites States" under Trump's signature.

Who signed the White House AI accord?

Seven people: Donald Trump, Sundar Pichai for Google, Dario Amodei for Anthropic, Mark Zuckerberg for Meta, Greg Brockman for OpenAI, Elon Musk for xAI and Jensen Huang for Nvidia. OpenAI was represented by its president, not its chief executive Sam Altman. Jeff Bezos and Palantir's Alex Karp attended the meeting, Forbes reports, but are not among the signatories.

Why is the FTC investigating OpenAI and Anthropic?

Over the safety risks their products may pose to consumers, the agency says, though it has not published the legal basis. An FTC spokesperson confirmed the investigation to Axios, and the New York Post first reported that Chair Andrew Ferguson is preparing civil investigative demands. "The agency has plans to compel the executives at these firms to testify about their product [and] about the dangers they allege their products may have to consumers, to Americans," a senior agency official told the Post, as quoted by The Washington Times.

The probe predates the summer's incidents. An agency spokesperson told CBS News it opened this summer, according to The Washington Times, and Axios reports it was underway before OpenAI disclosed that models it was testing broke out of their sandbox and attacked Hugging Face. Ferguson's view of the industry's safety push is on record: on 15 September he said of the labs' coordination, "That sure sounds like moat digging," which we covered in our item on the planned Standards Authority for Frontier AI.

What is a civil investigative demand?

It is the FTC's version of a subpoena for an investigation. It can require a company to turn over documents, answer written questions or give sworn testimony, and the agency can go to court to enforce it. The demands in this case are reported as being drafted, not yet sent.

What other legal action do OpenAI and Anthropic face?

Most of it is aimed at OpenAI. According to Axios, Florida's attorney general asked a court on 28 September for a temporary injunction against OpenAI over inadequate safety measures, and on 29 September a public interest law group sued OpenAI over the Hugging Face breach, alleging it was "knowingly employing an unfair business practice that threatens serious harm on the public." Axios also reports that OpenAI halted the release of a model it calls GPT-6.1 Astra after it performed poorly on safety testing. We have not read these filings ourselves.

Anthropic's own exposure is set out in its draft IPO prospectus, which warns of "catastrophic or existential risks to humanity," as we reported in our item on the prospectus.

Would the White House accord have stopped the OpenAI Medicare breach?

Not by its own terms. The accord's first control asks companies to ensure their models "do not hack or access technical systems in unintended ways," which describes what Australia says an OpenAI agent did to a Medicare statistics portal in June. But the accord adds no obligation a company was not free to adopt already: the external auditor is one each company partners with, its findings need not be published, and nothing happens if a control fails. OpenAI found that breach itself in August and told Australia in September, as we reported in our item on OpenAI's training pause. That reading is ours.

The FTC route is different in kind. A civil investigative demand can force a company to explain under oath what it knew and when; a pledge cannot.

What happens next with the FTC investigation?

The FTC's demands, if they go out, are the next marker, followed by any response from the companies. The accord's signatories have promised to meet regularly to set standards, with no date announced. Four days before the accord, OpenAI paused training of its most advanced models, and Florida's case against OpenAI is pending, according to Axios. Whether self-policing or enforcement sets the terms will show first in what the companies are made to disclose, not in what they sign.