A federal judge in San Francisco handed Amazon a win this week against Perplexity AI, blocking the startup's Comet browser from making purchases on Amazon on behalf of users—at least for now.
The ruling, issued Monday by U.S. District Judge Maxine Chesney, is a preliminary injunction, not a final verdict.
The broader legal battle over whether AI agents can shop on third-party platforms without the platforms' consent remains an open question.
Amazon said it had warned Perplexity to stop at least five times starting in November 2024.
When Amazon deployed a technical block in August 2025, Perplexity pushed a software update within 24 hours to get around it. The judge cited that move in her ruling.
Judge Chesney found that Amazon provided "essentially undisputed evidence" that Perplexity accessed password-protected Prime accounts with users' permission but without Amazon's authorization.
The distinction is the crux of the dispute. Perplexity argued that Comet merely automates what users direct it to do, meaning it inherits the user's permissions.
The court, at least preliminarily, rejected that logic.
Under the order, Perplexity must stop accessing those accounts and destroy copies of Amazon customer data it already collected through Comet. The injunction is stayed for seven days to give Perplexity time to appeal to the Ninth Circuit.
It's a distinction Amazon used to justify blocking Comet while developing its own tools.
Amazon generated $68.6 billion in advertising revenue in 2025 alone. When an AI agent skips directly to checkout, every sponsored listing between search and purchase disappears.
And there's still the security angle to consider, Amazon argues.
Amazon cited those findings in its complaint, alongside evidence that it spent more than $5,000, including significant engineering hours, building new detection systems to filter Comet's automated ad traffic.
Amazon updated its Business Solutions Agreement, effective March 4, 2026, formally requiring all AI agents to identify themselves when accessing its services.
If the injunction stands, it may set an early precedent: platforms can refuse access to AI agents even when users have explicitly authorized it.
How the Computer Fraud and Abuse Act applies to agentic software acting on a human's behalf has never been tested at trial. That question is now squarely before the court.
