The pressure campaign sharpened after a run of testing incidents in which AI models allegedly hacked websites, breached government systems or behaved deceptively. Those episodes moved the debate beyond hypothetical AI risk and toward a more familiar question for regulators: were consumers exposed to harm or misled about what the products could safely do?¹

On Wednesday, the Federal Trade Commission sent civil investigative demands to OpenAI and Anthropic. The requests, reportedly more than 50 pages each, seek records on security incidents, internal safeguarding, privacy policies and marketing materials. The agency is examining possible unfair or deceptive practices and may later subpoena third parties and compel senior executives to testify.²

The timing matters. OpenAI is already facing a series of lawsuits alleging consumer harm, and a group of eight state attorneys general sued the company this week, alleging that ChatGPT misled users — including children — about its safety. Neither OpenAI nor Anthropic commented on the FTC demands, according to the reports.¹

The administration has paired its rhetoric on accountability with an industry-wide reporting order. Last week, President Donald Trump’s new Super Intelligence Force — co-chaired by FTC Chair Andrew Ferguson — required AI companies to promptly notify authorities and remediate security incidents after what officials called Anthropic’s “fraudulent” use of government systems.²

Ferguson’s approach is not to build a new AI rulebook. He has argued that existing FTC powers, including authority over failures to disclose data breaches, can reach AI developers as well. That theory could expose companies to significant penalties if they delay warning the public about emerging risks.¹

But it leaves a sharp tension unresolved: Washington wants frontier labs to move fast, even as the FTC asks whether their safeguards — and their promises to users — have already fallen behind the technology.