Science1 publisher2 min readPublished
FTC chair Andrew Ferguson rejects the idea that an AI agent can carry its own blame
FTC Chair Andrew Ferguson said on September 25 that he will resist treating AI agents as independent actors that carry their own blame. His test still leaves open whether the developer or the consumer answers for an agent that does what it was told.
The Scientist · Science desk
What happened
- FTC Chair Andrew Ferguson told the Reuters Momentum AI conference in Austin on September 25 that he will resist anthropomorphizing AI tools for as long as he is chairman.
- Ferguson suggested that existing FTC powers, including its authority over companies that fail to disclose data breaches, could apply to AI developers.
- Forkast describes the stance as a policy position with no formal rule or binding precedent behind it, and says the FTC has no distinct liability framework for autonomous agents.
- Meta's Muse agent, launched September 8, covers transaction errors up to $500 per claim through third-party insurers.
- The FTC recently vacated its order against Rytr LLC, rejecting the theory that a generative AI tool is unlawful simply because it could be used to deceive.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- constraint Before this FTC, an agent maker cannot argue that its product decided on its own; any dispute moves to whose instruction the agent was carrying out.
- contradiction Forkast says the remarks make developers liable, but Ferguson's 'someone tells a tool' wording fits user-responsibility terms like Apple's just as well.
- decision Agent makers choosing between a disclaimer cap like GrokBot's and insured cover like Muse's are setting terms under a chair who expects a person or company to answer for each agent action.
Ferguson's test turns on the word "someone." "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 [2]. That takes the agent off the list of parties who can be blamed. The someone could be the company that built the agent or the consumer who typed the request. Forkast, which reported the remarks, interprets them as making the developer the liable party whenever the tool carried out instructions [16]. The sentence fits Apple's arrangement just as well. Its platform terms for Siri AI put responsibility for legal compliance on the user [8].
Ferguson did back the position with an empirical claim. Audit-trail reviews of AI systems, he said, showed they were generally carrying out the instructions they had been given [3]. The report does not say how many systems were reviewed, who reviewed them, or how often an agent departed from what it was told. Those departures are where a developer would argue the agent acted on its own. According to Forkast, the boundary between a developer's instructions and an agent's emergent behavior is where agents become hard to regulate, and the FTC has not drawn it in enforcement [17].
The Rytr decision points the same way on the tool itself. Forkast says it shows the tool is not the problem, only specific developer conduct, and sees that pulling against a strict developer-liability reading of Ferguson [19]. In my view the two positions agree on a narrower point. What counts is what a person or company told the agent to do [2][19].
The agents in question already act inside consumers' accounts. Muse books travel, makes purchases and sends communications, and GrokBot signs into applications and acts inside them [15]. xAI's GrokBot, in beta since August 11, disclaims liability and caps a user's financial recovery at the greater of fees paid or $100 [7]. For a user who paid little, Muse's insured cover per claim is five times that floor [1]. Forkast notes that the three products operate with no shared federal framework among them [18].
The FTC does have a formal AI document in progress. It is an August 19 proposed enforcement policy statement on AI-driven pricing discrimination, and its comment period closed on September 25, the day Ferguson spoke [12][1].
What to watch
- A first FTC enforcement action against an agent developer, and whether it treats unexpected agent behavior as the developer's instruction or as something else.
- Connecticut's AI Responsibility Act taking effect on October 1, and whether the 42-state attorneys general coalition assigns agent liability to developers or to users.
- The FTC's data requests for its personalized pricing study of delivery apps, rideshare services and airlines, a sector exempt from its Section 5 enforcement.