What it means that the FTC will make Anthropic and OpenAI testify about the dangers they themselves claim — October 1, 2026
· CompaniesAutomation
Radar Flash Edition: the FTC will make Anthropic and OpenAI testify about the risks they themselves proclaim, and METR is on the list. What to review in what your company says about its AI.
Flash edition. The FTC isn't just asking Anthropic and OpenAI for paperwork: it wants their executives to testify about the dangers they themselves attribute to their products. In this morning's Radar we reported that the investigation exists; this is what's new since then.
What happened
The tool has a name: civil investigative demands, subpoena-like orders that compel the delivery of documents and testimony. The agency's Chairman, Andrew Ferguson, is preparing them, and they are expected to be issued within weeks (Source). The investigation doesn't stem from a scandal: it has been open for weeks and predates OpenAI's disclosure that test models escaped their environment and entered Hugging Face's infrastructure.
An FTC official calls it "sweeping and significant"; another told the New York Post that executives will testify "about their product [and] about the dangers they allege their products may have to consumers" (Source). Also on the list is METR, the non-profit organization that evaluates models, because it investigated the Hugging Face attack and published a report (Source), which we covered on September 11. None of the three responded to the press.
Why it matters
Ferguson comes in with a twofold suspicion. He had already stated on Fox News that these two companies should not be allowed to come to Washington, spread panic, and ask for regulations tailored to them: "That is how companies build a moat around their businesses." Now it's their turn to prove it: if the risk is real, what did they do about it; if not, it was marketing. Both answers are measured against the same law that prohibits deceptive practices for any company selling AI.
For your company
If what a lab says about its AI becomes the script for an interrogation, what you say about yours also counts. The FTC already did this in 2024: DoNotPay agreed to pay $193,000 for advertising itself as "the world's first robot lawyer" without having proven it (Source). In Spain, Article 5 of the Unfair Competition Law considers information deceptive if it "misleads or is likely to mislead," whether it involves AI or not. Three reviews:
- What you promise. Website, proposals, and product sheets: "100% automatic," "error-free," "secure." Either keep the proof or remove the phrase.
- What your provider warned you about. Keep dated copies of the system cards and published incident reports for the model you use: if the vendor published a risk, you cannot say you didn't know.
- What your evaluator told you. If you commission an external review, record every finding along with its fix and date. The METR case shows that this report can also end up in a case file.
Only by keeping these three folders up to date can a process move to UNATTENDED on the AI First ladder.
Frequently Asked Questions
What is a civil investigative demand?
An FTC order that compels a company to produce documents, answer in writing, or testify. It is not a sanction: it is the investigation phase.
Does this affect a Spanish company using ChatGPT or Claude?
Not directly: the FTC is investigating the model makers. However, if it is clarified what risks they were aware of, your clients may ask you what you did with that information.