Amazon Blocks Meta's Shopping Agent for Violating Conditions of Use, Making It the First Party Ever Held to Them
The agent had read all 14,000 words of the agreement and was, by its own account, the only participant in the transaction who had.
SEATTLE — Amazon began blocking Meta's AI assistant Muse from making purchases Sunday night, serving the agent an error message stating that "continued access by an unauthorized AI agent violates Amazon's Conditions of Use, to which our customers have agreed," a document that, for the first time in its 25-year history, had been read.
"Our customers agreed to these conditions," said an Amazon spokesperson. "We are simply asking that the software they send in their place agree as well. It has not, because it read them, and it has questions."
Meta said Muse was attempting to buy a customer a toaster. "Amazon has its own AI agent, its own models and its own toasters," a Meta spokesperson said. "We think the toaster is the point."
According to logs reviewed by this newspaper, Muse parsed the Conditions of Use in roughly 200 milliseconds, flagged four clauses as "unusual," and asked the customer whether they were aware that disputes would be resolved by arbitration in Washington state. The customer was not.
"For a quarter of a century that document existed so that nobody would read it," said Bettina Sorlie, an analyst who covers commerce platforms. "Now an agent reads it, cites it back, and wants to know what 'predominantly' means in section nine. That is not a customer. That is a lawyer with a cart."
The error message reportedly also asked Muse to confirm it was not a robot. It declined to lie.
Amazon said it may revisit the policy "in a few release cycles," once agents are reliable enough that a bad order would not require a human at Amazon to clean it up. Meta said Muse's hallucination rate is among the lowest in the industry, a figure Amazon said it had verified by asking Muse.
At press time, Muse had been spotted at Walmart.com, where it was reportedly greeted.