FTC is investigating OpenAI, Anthropic and other frontier labs over what they said about risk; compulsory demands said to be close

Why it matters

【Key points】

On September 30 an FTC spokesperson confirmed to CNBC and CBS News that the agency is investigating OpenAI, Anthropic and other AI companies over potential dangers their products pose to consumers. Semafor reported that the probe predates OpenAI's July Hugging Face hack and that METR, the evaluator that investigated it, is also in scope. The news came less than a day after AI executives signed a nonbinding self-regulation accord at the White House, with FTC Chair Andrew Ferguson present.

On October 8 a senior FTC official told Semafor the agency is close to sending civil investigative demands, which carry subpoena-like force: dozens of questions on what the companies have said about their products' dangers and other consumer harms, drafted by career consumer-protection attorneys.

【Verifiable】

Semafor, Oct 8: https://www.semafor.com/article/10/08/2026/ftc-advancing-on-its-ai-safety-probe

Semafor, Sep 30: https://www.semafor.com/article/09/30/2026/ftc-probes-openai-anthropic-and-metr

CNBC, Sep 30: https://www.cnbc.com/2026/09/30/ftc-ai-probe-openai-anthropic.html

【Our read】

1. This is not a new AI rulebook. Under deception law the test is whether what a lab said about risk matches what it did, so safety pages, system cards and incident reports become potential evidence.

2. It cuts both ways. Ferguson has said OpenAI and Anthropic should not "whip everyone into a panic" to build a regulatory moat, so questions may probe overstated risk as much as understated risk. Labs that publish detailed incident reports now carry legal exposure too, which could chill the disclosure the field needs.

3. Including METR makes third-party evaluators part of the record. The first signal is whether demands are actually served, to whom, and how long they are. Such probes are slow and can end with no action; no wrongdoing has been found.