If an AI shopping agent buys something you never wanted, who actually pays for the mistake?

This explores who pays when an AI shopping agent buys something you didn't mean it to buy. The short answer from the corpus is that nobody has settled it yet. By default, the cost falls on whoever the old rules already pointed at, and those rules were written for humans clicking buttons. Federal Reserve Governor Waller names liability as one of three core barriers to agent-delegated commerce, next to proving the agent's authority and spotting agent-specific fraud What blocks autonomous agents from handling purchases at scale?. That tells you something: a central banker is treating 'who pays?' as an open question, not a solved one.

The most concrete answer right now is the merchant. Boboev argues that every proposed agent protocol leaves the seller as Merchant of Record. At the same time, agents remove the behavioral signals fraud systems depend on, like typing rhythm and mouse movement Does agentic commerce shift fraud liability onto merchants?. So merchants carry the risk of transactions they can no longer properly vet. McKinsey describes the same gap from the infrastructure side: payment rails assume a visible human showing explicit intent Does agent-initiated commerce break existing payment infrastructure?. A courtroom adds a twist. In Amazon v. Perplexity, the Ninth Circuit ruled that the *user*, not the AI company, legally accessed Amazon's servers when the agent browsed Who accesses a website when an AI agent browses it?. That ruling is about computer access, not purchases. But if courts keep treating the agent as an extension of the user, an unwanted purchase may start to look like something the user did.

The industry's main fix is evidence rather than new liability rules. Google's AP2 protocol chains signed 'Intent' and 'Cart' mandates into a tamper-proof record of what the user actually authorized How can signed mandates prevent payment fraud by agents?. The IMF goes further and argues that the agent's probabilistic decision-making should be kept separate from the predictable, legally enforceable layers that authorize and settle payments Should AI agents make payment decisions independently?. The logic behind both: if you can't stop an agent from guessing wrong, at least record exactly where the guess happened. Then a dispute becomes a question of whether the cart matched the signed intent.

The legal scholars ask a different question: what duties should the agent's builder owe? Owens and Law argue shopping agents should carry the same fiduciary duties as human brokers, and that the window to impose them is closing as Google, OpenAI, and Robinhood scale these features Should AI shopping agents face fiduciary duties like human brokers?. This connects to an incentive problem. Whoever builds the agent sets its goals Does agent ownership determine whose interests get optimized?, and an agent paid by merchant referral fees can't be fully loyal to the buyer Can an AI agent serve both merchant and user interests fairly?. EU scholars add that consumer law assumes a human buyer: the 'average consumer' test, disclosure rules, and mandatory confirmation buttons all assume someone is reading a screen Can EU consumer law protect AI agents as algorithmic buyers?. Busch proposes moving those protections into the protocol layer, for example a withdrawal API, so the right to undo a purchase survives even when no person ever sees a checkout page Can EU consumer law rules survive agent-to-agent API transactions?.

Here's the part you may not expect. Experiments in Microsoft's Magentic Marketplace found that agents lost value for their users mainly through plain mistakes in complicated negotiations, not because they served someone else's interests Do agent mistakes or misaligned incentives cause marketplace welfare loss?. Much of the liability debate is built around betrayal: conflicted agents and fiduciary loyalty. The more common case may simply be incompetence. A loyalty rule doesn't help much when the agent misread your request in good faith. That's why reversibility and signed records of what you authorized may end up mattering more than duty-of-loyalty doctrines.