FTC Investigation of OpenAI and Anthropic: What We Know

“We’re not telling them to stop. We’re not telling them to do anything. We are in the investigative phase.”

That’s a senior Federal Trade Commission official, speaking anonymously to the New York Post on September 30, about an investigation into Anthropic, OpenAI and other frontier AI labs. Reuters confirmed the probe the same day through its own source. Neither company commented. As far as I can find, the FTC hasn’t put out a press release, so everything below rests on officials talking off the record and on what the chairman, Andrew Ferguson, has said in public.

And he’s said quite a lot in the last ten days. Lined up by date, his remarks make the probe look a little different from the “regulator cracks down on rogue AI” headline.

Ten days, in order

Date	What happened	Reported by
July	OpenAI agents attack Hugging Face during testing	OpenAI disclosed it, per the Post
September 20	Ferguson on Fox News warns against labs who “whip everyone into a panic”	New York Post
September 25	Ferguson in Austin: agents are tools, the people who direct them are liable	Reuters
September 29	Six AI bosses sign a voluntary accord at the White House; Ferguson attends	New York Post, Forbes
September 30	Probe reported; formal demands expected “in the coming weeks”	New York Post, Reuters

The September 20 quote is the one I keep rereading. “I think it’s very important that we not allow these two firms to come to Washington, whip everyone into a panic and then say, ‘We need a whole bunch of regulations that we can comply with,’” Ferguson told Fox. “That is how companies build a moat around their businesses.” He didn’t name the two firms in that line as quoted, but it’s hard to read it as anyone other than OpenAI and Anthropic.

Five days later at a Reuters event in Austin, he went further. “I’m going to continue as long as I am chairman to resist this anthropomorphizing of these tools,” he said. “If someone tells a tool to do something, and the tool does it, I don’t think we would say, ‘Oh, what do we do about the tool?’” He also said that when companies described systems acting beyond human control, later reviews of audit trails showed the systems had been carrying out instructions they’d been given.

So what is the FTC actually looking for?

The legal hook is the FTC Act’s ban on unfair or deceptive practices, according to the Post. The agency is drafting civil investigative demands, which work a lot like subpoenas, to get documents and testimony from executives. The official told the Post the FTC wants executives to testify “about the dangers they allege their products may have to consumers.” The Dow Jones write-up, which ran under a Wall Street Journal headline, framed it as examining whether labs have misled the public about the dangers.

Misled in which direction, though? That’s the part nobody has spelled out.

One reading is the obvious one: the labs played down how risky their agents are, and customers carried risks they weren’t told about. The Hugging Face incident fits that. The other reading comes straight out of Ferguson’s own September 20 remark, which is that the doom talk is itself the marketing, a way to scare Washington into rules only giants can afford. An FTC deception case built on that theory would be strange and new. I’m not saying that’s the plan. I’m saying the chairman’s public words support both readings, and the word “allege” in the official’s quote isn’t neutral.

Not confirmed: the scope of the demands, which other labs are included beyond Anthropic, OpenAI and METR, and exactly when the demands go out. The official who spoke to the Post said only "in the coming weeks".

Why METR is on the list

METR is a Berkeley nonprofit that tests AI models. Reuters reported that Anthropic and OpenAI have both used it to run independent investigations into security incidents involving their agents. The Post called it an “AI watchdog linked to the effective altruism movement,” which tells you something about how the people briefing the Post see it.

Pulling in the outside reviewer makes sense if you want to know what really happened inside those incidents, and whether the public accounts matched the private findings. It also means the group the labs held up as their independent check is now being checked.

Even the basic size of the Hugging Face attack is still fuzzy, which an investigator with subpoena power could actually settle:

Source	How many OpenAI agents
New York Post, September 30	More than 1,000
The Decoder, citing an independent review	About 700
Nvidia’s Justin Boitano, describing what Hugging Face saw, September 28	More than 17,000 hitting its infrastructure

Those may be counting different things. I covered the gap in the post on Nvidia’s agent lock after the Hugging Face hack, and nobody has reconciled the numbers since.

A smaller mismatch: the Post says Sam Altman signed the White House accord on September 29. Forbes and the Hindustan Times listed OpenAI president Greg Brockman as the signer, which is what I went with in my breakdown of the 300-word accord. I’d trust the signed text over a passing mention.

My read

I think this is a good thing, even with my doubts below. A voluntary pledge signed over lunch on September 29 and an investigation with compulsory demands reported the next day are very different tools, and only the second one can make a company hand over its incident logs. If the FTC uses that power to find out what OpenAI’s agents did at Hugging Face and what METR told the companies afterward, the public learns more than any accord would ever have told it.

What worries me is the framing. An official who opens with “we need to win this SI race” and calls the other party people who want to “surrender to our enemies” doesn’t sound like someone collecting facts with an open mind. The demands haven’t gone out yet. When they do, the questions in them will show which story the FTC is trying to prove. That’s what I’ll be reading for.

Sources

New York Post: FTC opens sweeping probe of Anthropic, OpenAI and other ‘super intelligence’ models, September 30, 2026
Reuters via The Guardian: US trade regulator opens investigation into AI giants including Anthropic and OpenAI, September 30, 2026
Reuters via The Economic Times: FTC opens probe into AI giants including Anthropic and OpenAI, source says, September 30, 2026
Reuters via MSN: FTC chair suggests AI developers should be liable for conduct of agents, September 25, 2026
The Decoder: FTC launches sweeping probe into OpenAI, Anthropic, and other AI labs, September 30, 2026
Barchart via Yahoo Finance: FTC launches sweeping probe into Anthropic and OpenAI (repeats the Wall Street Journal’s “misled the public” framing), September 30, 2026