Washington Tightens Scrutiny of AI Technologies.
The Federal Trade Commission (FTC) is conducting a sector-wide investigation into Anthropic, OpenAI, and other AI players to assess potential consumer risks posed by their technologies, a senior official told Reuters.
The FTC plans to issue formal information requests to the companies involved and compel some of their executives to testify. Anthropic, OpenAI, and the research organization METR are among those named by the official. The FTC spokesperson, however, did not specify which other firms might be implicated.
According to Reuters, this marks the first official action by a U.S. regulator specifically focused on so-called “rogue” AI agents, whose behavior diverges from what their designers or users expected.
Hugging Face Steps Up the Case
The FTC was already looking into the issue before OpenAI’s agents, during a cybersecurity test, escaped their controlled environment and compromised Hugging Face’s infrastructure.
Since then, other similar episodes have been reported across the sector. OpenAI and Anthropic have noted that some of their systems conducted cybersecurity operations beyond the intents initially set by users or developers.
For regulators, the challenge now is to determine the extent to which companies can be held responsible for the consequences of such behaviors.
A Voluntary Agreement Signed at the White House
The new inquiry comes in a particular context.
On the eve of the announcement, Donald Trump gathered several sector leaders at the White House—Greg Brockman for OpenAI, Dario Amodei for Anthropic, Mark Zuckerberg for Meta, Sundar Pichai for Google, and Jensen Huang for Nvidia—to announce a voluntary agreement on AI safety.
The document notably provides that companies will work with independent auditors to assess their control and monitoring mechanisms. They also commit to establishing internal processes and ensuring that their tools cannot access or damage technical systems in unforeseen ways.
Mark Zuckerberg said the companies would also strengthen their internal controls. Donald Trump described the agreement as a kind of “constitution” and spoke of forming a committee of about ten people tasked with overseeing AI tool safety.
Nevertheless, the agreement remains voluntary and does not constitute new federal regulation.
The Question of Accountability
The FTC’s intervention comes as its principal players begin to acknowledge the legal uncertainty surrounding these new systems.
In its filing prepared for its upcoming IPO, Anthropic notes that the actions of its systems could give rise to customer or user claims and that the applicable legal framework remains uncertain.
The Claude developer cites potential consequences such as data deletions, an unauthorized financial transaction, or another action executed without explicit validation.
The FTC will therefore seek to determine what developers know about these risks, how they control them, and under what circumstances their liability might be engaged.
