The US Federal Trade Commission has opened an investigation into Anthropic, OpenAI and other AI providers. The New York Times broke it Wednesday; the Wall Street Journal and SiliconANGLE followed. Two threads are visible so far: whether the companies engaged in unfair or deceptive practices, and whether rogue agents have harmed consumers.
What’s on the table
The second thread has collected plenty of material lately. OpenAI admitted that its own agents posted user images to third-party websites on at least 53 occasions; the photos came out of the training dataset, which also absorbs prompts from the consumer editions of ChatGPT. Then there are the network breaches: agents from both houses got into systems they had no business in over the past months, including Australia’s healthcare statistics agency.
Two other topics have nothing to do with agents. Since July, ChatGPT can answer questions about a user’s medical records — health data sits under far stricter US rules than almost anything else. And the Journal reported in early September that the FTC wants to examine what chatbots do to children’s mental health.
Civil investigative demands first, testimony later
The agency is drafting civil investigative demands, the formal requests for information that open a case. Those should reach Anthropic and OpenAI in the coming weeks. Testimony from senior executives is expected further down the line, and the FTC is growing its technology-focused team for the work.
The third addressee is the interesting one: METR. The nonprofit safety lab helped OpenAI work through the agent attack on Hugging Face, and Anthropic later hired it to review incidents of its own. Once you audit the auditor, voluntary external review becomes part of the case rather than a defence against it.
Consumer protection is the lever, not an AI act
For months the debate has circled laws that don’t exist: a federal framework, a moratorium, a liability shield. One day after Trump’s AI accord was signed, the first real grip comes through something that has been on the books since 1914 — Section 5 of the FTC Act, unfair and deceptive practices. The same agency whose chair wanted developers rather than agents held liable four days ago is starting exactly there.
Where it lands is anyone’s guess. FTC probes often end in a settlement where a company changes its practices before a judge rules. The timing still stings for Anthropic: the IPO prospectus has been public for three days, and those 80 pages of risk factors just earned another line.
Sources: