Court lets Perplexity’s AI agent shop on Amazon, overturning a ban

An AI agent can shop on Amazon whether Amazon likes it or not. A US appeals court has overturned an injunction that barred Perplexity’s shopping agent from the retailer, handing an early win to the software that browses and buys on people’s behalf.

The Ninth Circuit ruled that Amazon’s core legal claim did not hold. Because Perplexity’s Comet agent acts only when a user tells it to, the court found it is the user, not Perplexity, who is accessing Amazon’s servers.

Comet is the agent built into Perplexity’s AI browser. Point it at a product and it can navigate Amazon and complete a purchase, the kind of autonomous errand a wave of AI companies are betting will replace ordinary online shopping.

The retailer argued the agent amounted to unauthorised access, said Perplexity had disguised Comet as a normal Chrome browser to avoid detection, and pointed to a 2024 agreement in which Perplexity had agreed to pause agentic shopping.

Perplexity cast the fight as one over who controls a person’s shopping experience, arguing that users should be free to bring whatever software they like to websites they already have the right to visit.

The dispute turned on a decades-old hacking law. Amazon leaned on the Computer Fraud and Abuse Act, written for a very different kind of intrusion, and the court decided that a tool doing a user’s bidding does not fit its definition of unauthorised access.

The reasoning is the real prize. If an agent acting on a user’s instruction is legally the user, then platforms lose one of their sharpest tools for keeping third-party agents out, and the door to agentic commerce swings open.

For Amazon, that is precisely the fear. It has spent years designing a shopping experience it controls, from search ranking to recommendations, and an agent that skips all of it strips the retailer of the influence, and the ad revenue, that layer generates.

Amazon has been building its own agentic layer, putting Alexa inside the search bar as it races to own the shopping agent rather than be bypassed by someone else’s.

The shift is already well underway elsewhere. China’s biggest platforms have gone furthest, replacing the search bar with AI agents that shop for users, a preview of where Western retail may be heading.

The mechanics are converging too. Alibaba has wired its own model straight into Taobao for end-to-end agentic shopping, the sort of first-party integration Amazon would prefer to a rival’s bot crawling its aisles.

The fight is not over. Amazon said it remains confident and is weighing its options, which include seeking a rehearing or asking the Supreme Court, while the underlying lawsuit continues in a San Francisco court.

But an appeals-court ruling carries weight. It gives other agent-makers a template to argue that acting for a user is not hacking, and it puts platforms on notice that blanket bans may not survive contact with the law.

It also nudges a bigger, unsettled question. When an agent does something harmful rather than helpful, it is still unclear who is liable, and treating the agent as the user has consequences well beyond a shopping cart.

The stakes are commercial as much as legal. Retail media has become one of Amazon’s fastest-growing profit lines, and an agent that ignores sponsored placements and buys the cheapest match threatens the ad economics holding much of that business together.

The deeper tension is structural. Websites were built for humans clicking through pages, and a bot that reads and buys at machine speed strains both the technical assumptions and the business models underneath them.

For now, the agents have the upper hand. A court blocked the web’s biggest retailer from keeping them out, in a ruling that may prove much bigger than it first appears.