At the Reuters Momentum AI conference in Austin on September 25, Federal Trade Commission Chair Andrew Ferguson offered the clearest federal position yet on a question that consumer-facing AI products have made urgent: who pays when an autonomous agent makes a purchasing decision, books a trip, or sends an email that causes harm?
His answer was unambiguous — and it carries direct consequences for the companies now shipping real agents to consumers.
I’m going to continue as long as I am chairman to resist this anthropomorphizing of these tools. 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?’
The remarks reject the emerging “autonomous actor” defense — the argument, advanced implicitly by some agent developers, that AI systems with sufficient autonomy should be treated as independent decision-makers rather than instruments of their creators. Ferguson said audit trail reviews of AI systems showed they were generally carrying out instructions they had been given, and suggested existing legal tools — including FTC authority against companies failing to disclose data breaches — could apply to AI developers.
This is a policy position, not a formal rule or binding legal precedent. There is currently no distinct FTC liability framework for autonomous agents. But the statement signals the direction of federal enforcement under Ferguson’s leadership, and it arrives at a moment when the products making it urgent are already in consumers’ hands.
The Agent Products That Make This Concrete
Two weeks before Ferguson’s remarks, three consumer-facing AI agents shipped under three different liability models — and none operated under a shared federal framework. Meta’s Muse, launched September 8, offers insured purchase protection through third-party insurers, covering transaction errors up to $500 per claim. xAI’s GrokBot, in beta since August 11, disclaims liability entirely, capping financial exposure at the greater of fees paid or $100. Apple’s Siri AI relies on traditional platform terms that place legal compliance responsibility on the user.
These are not hypothetical products. Muse books travel, makes purchases, and sends communications on behalf of users. GrokBot signs into applications and acts inside them. The question Ferguson addressed — whether the developer or the agent itself bears responsibility for what happens — is already a daily consumer reality.
Ferguson’s answer narrows the space for any developer who might argue their agent acted independently. If the tool carried out instructions, the developer is the liable party.
The Rytr Contradiction
Ferguson’s position carries an internal tension worth noting. The FTC recently vacated its order against Rytr LLC, expressly rejecting the theory that a generative AI tool is unlawful simply because it could be used to facilitate deception. That decision pulls in the opposite direction from a strict developer-liability reading: it says the tool itself is not the problem, only specific developer conduct.
Ferguson’s remarks attempt to square this circle by focusing on what the developer told the agent to do, rather than on the agent’s general capabilities. But the boundary between “the developer’s instructions” and “the agent’s emergent behavior” is exactly where autonomous agents become difficult to regulate — and the FTC has not yet drawn that line in enforcement.
The Airline Gap
Ferguson also announced that the FTC is preparing to request data from consumer-facing companies for a market study on personalized pricing, identifying delivery apps, rideshare services, and airlines as his primary areas of concern.
But airlines are explicitly exempt from FTC Section 5 authority under 5(a)(2) of the FTC Act. The agency can study the sector but cannot enforce there. This creates a structural blind spot in Ferguson’s stated priorities — one the industry will note.
The personalized pricing inquiry connects to the FTC’s broader trajectory: from the August 19 proposed enforcement policy statement on AI-driven pricing discrimination (comment period closed September 25), through the 42-state AG coalition building agent liability frameworks Congress won’t, to Connecticut’s AI Responsibility Act taking effect October 1.
What Comes Next
Ferguson’s statement is enforcement by signal — a clear direction without a formal rule. Companies building autonomous agents now know the FTC chair’s framework: the developer is the actor, the audit trail is the evidence, and the “autonomous actor” defense is not one the agency will entertain.
For Muse, GrokBot, and the next wave of consumer agents shipping this fall, the practical question is narrower than the philosophical one. It is not whether AI agents have agency. It is whether the company that shipped the agent can show, through its audit trail, that the agent did what it was told to do — and whether that showing will be enough when something goes wrong.