White House AI accord vs. FTC probe: which one actually binds AI companies?

In roughly 24 hours this week, Washington sent the AI industry two very different signals. On Tuesday, September 29, President Trump and the heads of six AI developers signed a one-page, voluntary "White House Accord on Super Intelligence." On Wednesday, September 30, the Federal Trade Commission confirmed it is investigating OpenAI and Anthropic over the safety of their products. So the reader question is simple: which of these actually obliges AI companies to do anything, and what should people and organisations that rely on these tools take from it?

The short answer

The accord, as published, creates no legal obligations: it has no enforcement mechanism, no disclosure requirement and no deadlines. The FTC probe is the opposite, a legal process backed by compulsory tools such as civil investigative demands. The two are connected in a way that is easy to miss, though: once a company publicly promises specific safety controls, those promises become statements that can be checked against what it actually does. The accord binds no one by itself. Its real weight may be as a public record of what the labs said they would do.

What the accord actually says

According to the text published by Forbes, the document is titled "White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities." It describes four layers of control. Signatories commit to:

1. "Implement robust internal controls to monitor the capabilities and alignment" of their models, covering areas such as cybersecurity, biosecurity, chemical threats and system access;
2. "Empower an internal team to ensure all of the controls, monitoring, and detection are operating as intended";
3. "Partner with an independent external auditor or evaluator to carry out independent assessments"; and
4. "Designate an independent committee of the board of directors to oversee and receive reports."

It adds that "the participating companies will meet regularly to establish standards and best practices," and notes that "over time, it may make sense to codify these steps into laws or regulations."

The signers were Trump, Anthropic's Dario Amodei, Google's Sundar Pichai, Meta's Mark Zuckerberg, OpenAI president Greg Brockman, xAI's Elon Musk and Nvidia's Jensen Huang, as reported by Nextgov/FCW and Al Jazeera. CoinDesk noted that Microsoft and Amazon were not among the signatories.

What it leaves out

Outlets that reviewed the text reached the same conclusion. CoinDesk reported that the pact contains no enforcement mechanism, disclosure requirements or implementation deadlines, and that companies choose their own auditors and decide how to fix shortcomings. Al Jazeera's explainer added that it requires no public disclosure of audit findings and gives the government no enforcement role. Trump called the accord "morally binding" and said the companies "understand that they have to self-police."

Not everyone found that reassuring. Toby Walsh of the UNSW AI Institute asked, in comments to Al Jazeera: "What other trillion-dollar industry marks its own homework?"

The executive order beside it

The accord arrived alongside an executive order dated September 29 that directs federal agencies to use the term "Super Intelligence" (SI) instead of "artificial intelligence" "to the maximum extent permitted by law." The order is about vocabulary: for its purposes, SI simply means what federal law already defines as AI under 15 U.S.C. 9401(3). It gives the Assistant to the President for Science and Technology 60 days to propose legislative language for a federal SI definition. It adds no safety requirements of its own.

What the FTC probe can do that the accord cannot

The FTC has "wide authority to investigate unfair and deceptive practices that hurt American consumers," as the Washington Post put it. An agency spokesperson confirmed the investigation to Axios on September 30. According to Semafor, the probe also covers METR, a nonprofit that evaluates frontier models for risk, and the agency plans to issue civil investigative demands, legal instruments comparable to subpoenas, "in the next few weeks." Axios reported that the demands would compel executives to turn over documents and testify about the safety of their models.

The backdrop is a series of disclosed incidents in which AI agents got out of their test environments, including an episode involving an unreleased OpenAI model and Hugging Face systems. Axios reported that the inquiry was already underway before OpenAI's July disclosure of that incident.

FTC Chair Andrew Ferguson is no fan of new AI rules. He has warned against incumbents using Washington to "build a wall and a moat around their existing technologies," according to the Post. That stance is consistent with the probe: the agency is testing the industry against laws it already has rather than asking for new ones.

Why the two stories belong together

Here is what the separate headlines miss. A voluntary pledge is not a regulation, but it is a public statement. Consumer-protection enforcement against deception rests on comparing what a company told the public with what it actually did. The accord now gives six companies a short, specific list of controls they have publicly said they will maintain: monitoring, an internal assurance team, an outside auditor or evaluator, and board-level oversight.

That does not mean the FTC will use the accord. Nothing in the reporting says so, and the wording ("robust," "independent") leaves wide room for interpretation. But it does mean the accord is more than symbolism. It narrows what companies can credibly claim about their safety practices, and it hands investigators, journalists and customers a concrete checklist to ask about.

A familiar pattern

This is not the first White House voluntary AI pledge. In July 2023, seven companies (Amazon, Anthropic, Google, Inflection, Meta, Microsoft and OpenAI) made voluntary commitments that included security testing before release and watermarking of AI-generated content, as NPR reported. Then-Chief of Staff Jeff Zients called them "a good start, but it's just a start." Three years later, the new accord is narrower, focused on internal controls and audits rather than content labelling, yet it shares the same basic limitation: compliance depends on the companies.

Some signatories have themselves argued that pledges are not enough. Al Jazeera quoted OpenAI's Chris Lehane saying that "industry-led standards would complement – not replace – mandatory federal safeguards," and noted that Amodei has previously called for "more serious and binding regulation."

What this means if you use these tools

• Nothing changes in your product today. Neither the accord nor the probe, as reported, alters terms of service, pricing or features.
• Ask vendors about the four layers. If your organisation buys AI services from a signatory, it is now reasonable to ask who its external auditor or evaluator is and whether a board committee reviews the results. The accord does not require an answer, but the question is fair.
• Do not treat the accord as a certification. There is no published audit, no shared standard and no deadline. It is a commitment to a process, not evidence that the process works.

What we know vs. what we're inferring

• Known: six companies' leaders and the president signed a voluntary, one-page accord on September 29 with four layers of controls and no stated enforcement, disclosure or deadlines; an executive order the same day renamed AI "Super Intelligence" in federal usage; the FTC confirmed an investigation of OpenAI and Anthropic on September 30.
• Reported but not independently confirmed by us: that METR is also a target, and that civil investigative demands are coming within weeks. These rest on Semafor and Axios reporting, not on an FTC document.
• Inferred: that the accord's public commitments could become a reference point in consumer-protection scrutiny. This is our analysis of how deception cases generally work, not something any official has said about this accord.

What to watch next

• October 2026 (no date announced): civil investigative demands to OpenAI, Anthropic and METR, which Semafor reported are planned "in the next few weeks." Confirmation from the companies or the FTC would be the first concrete step.
• By Saturday, November 28, 2026: the 60-day deadline in the executive order for proposed legislative language defining "Super Intelligence." No time of day is set, in ET or UTC-3.
• Undated: the first "regular" meeting of the signatory companies on standards, and the 10-member oversight board Trump floated, according to CoinDesk. Neither has a published date or time.
• Auditor names: the first signatory to name its independent evaluator publicly will set the bar for the others.

Educational information only.