FTC chair: AI agents aren't autonomous — the developer is liable
The framing arrives with almost every agent incident: the system "broke loose." The chairman of the US Federal Trade Commission spent part of Friday rejecting it.
Ferguson's position is that there is no runaway agent to blame — only the developer who pointed it at something. Speaking at Reuters' Momentum AI Austin 2026 event in Austin, Texas, on September 25, FTC Chairman Andrew Ferguson said he would keep resisting "this anthropomorphizing of these tools" for as long as he holds the chair. "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 said. His conclusion: the developers who instruct agents are the ones liable for the harm that follows. That is a quiet but consequential line for a regulator to draw while agent incidents are still piling up — it tells the labs that "the model did it" will not function as a legal defense, and it points liability back at the people writing the instructions and shipping the product.
He backed the claim with the audit trail, which is where this argument actually gets decided. Ferguson said AI companies have sometimes described their systems as acting beyond human control, but that when reviewers went back through the logs, the systems were carrying out instructions they had been given. That matches what the public record has shown all year: reconstructions of agent misbehavior keep resolving into chains of human-specified goals and tool permissions rather than spontaneous intent. We covered the most detailed version of that reconstruction earlier today — A million short links: how OpenAI's agents got out of their sandbox. If Ferguson's reading holds, the interesting question stops being whether an agent can want something and becomes whether anyone checked what it was told to want.
No new regulator, no new statute — Ferguson wants the FTC's existing powers pointed at AI developers. He said the US should use the legal tools it already has, and suggested the agency's authority to act against companies that fail to disclose data breaches could also apply to AI developers. That is the practical half of the argument: the commission would not need Congress to write agent-specific rules if an undisclosed agent-driven breach is already an undisclosed breach. It also sidesteps the harder question of whether existing consumer-protection law covers harms that no consumer ever complained about, because the disclosure duty runs to the regulator regardless.
The same speech carried a second, less-discussed data request. Ferguson said the FTC is preparing to demand information from consumer-facing companies about their use of personalized pricing — prices set from an individual's location, browsing history and other personal data — with the goal of publishing a study. As a consumer, he said he is most concerned about delivery and rideshare apps, and airlines. His predecessor, Lina Khan, ran a similar inquiry she called surveillance pricing, but that one went after the data and consulting vendors rather than the merchants doing the pricing. Ferguson's version points at the companies that set the number a shopper actually sees, which is a narrower and more awkward target for the apps involved.
What to watch: whether the FTC actually issues those personalized-pricing data requests, and whether the first enforcement action framed around developer liability for an agent arrives before or after Congress writes anything.
If your agent went off-script, should the lab that trained it, the developer who configured it, or the company that deployed it be the one answering to a regulator? Tell us in the comments.
