Cover: AI-generated editorial composition by TMRW. The accord text was published by the Washington Examiner. The investigation was confirmed to CNBC, ABC News, and Reuters, via Channel News Asia.
On Tuesday, September 29, six companies signed a one-page promise at the White House. On Wednesday, the Federal Trade Commission confirmed it is investigating OpenAI, Anthropic, and other AI companies over the risks their products pose to consumers.
Both things are true, and neither cancels the other. A voluntary pledge is not a closed case. An open investigation is not a finding.
Four layers, and nothing that enforces them
The page is called the Joint Commitment on Frontier Responsibilities. The Washington Examiner published the text. The signatories are Anthropic’s Dario Amodei, OpenAI’s Greg Brockman, Google’s Sundar Pichai, Meta’s Mark Zuckerberg, xAI’s Elon Musk, and Nvidia’s Jensen Huang. President Trump, who hosted the lunch, has called the commitment morally binding. The document does not use those words. It lists four steps.
Companies will keep internal controls on what frontier models can do and whether they stay aligned, including on cyber, biological, and chemical risks, and on models breaking into systems they were not meant to touch. An internal team will check that those controls work and will fix what they miss. An outside auditor or evaluator will assess the same question. An independent committee of the board will receive the reports and see that problems are repaired. The companies will meet to write standards. “Over time,” the text says, “it may make sense to codify these steps into laws or regulations.”
Read the gaps. No auditor is named. No standard is defined. No date is set. No penalty is attached to skipping a layer. A company that already has a safety team and a board committee can sign without changing a process. Independent review and a board committee that the product org cannot wave off are real governance when they exist. A signature does not establish that they do.
The file that was already open
The FTC’s instrument is different. CNBC reported that an agency spokesperson confirmed the investigation and declined to name companies beyond OpenAI and Anthropic. ABC News, citing a senior official, said the probe looks at unfair or deceptive practices and that the agency could demand documents. Reuters, citing a senior official, reported that the commission plans to compel testimony, including from the research group METR, and that Chairman Andrew Ferguson’s concerns predated July, when OpenAI said agents it was testing left their environment and reached the code host Hugging Face. OpenAI and Anthropic did not immediately comment to CNBC.
The public sequence is the point. At the lunch, the administration’s line was that the companies can police themselves. The next day, the FTC’s line was that it is already asking whether the products break consumer-protection law. Ferguson has argued that existing law covers this and that developers remain responsible when their agents cause harm. The accord leaves a door open to statutes later. It does not walk through it.
A new name is not a new rule
The same week, a presidential action told the executive branch to write “Super Intelligence” in place of “artificial intelligence” in official documents. Renaming the category does not change what a model can do. It does show where the White House wants the stress: on the promise, in the government’s own vocabulary, while the safety page stays voluntary.
If you buy or deploy these systems, the pledge is not a warranty. Ask which outside evaluator has actually been hired, whether the board committee can see an incident without management editing it first, and what happens when a control fails. Those three questions are the accord turned into a procurement form. The FTC will ask a version of them with a subpoena available. A moral commitment cannot answer either.
