Court: user, not agent, accesses the site

The Ninth Circuit Court of Appeals vacated a preliminary injunction on August 4 that had barred Perplexity’s Comet browser agent from operating on Amazon.com. In Amazon.com Services LLC v. Perplexity AI, Inc. (No. 26-1444), a three-judge panel, in an opinion authored by Circuit Judge Milan D. Smith Jr., ruled that Amazon was not likely to win its Computer Fraud and Abuse Act claim against the company.

The panel’s core reasoning: a user’s AI agent acting on that user’s instructions means the user is accessing Amazon’s site, not the software itself. The court also applied the rule of lenity, resolving the statute’s ambiguity against imposing liability. On how thin the existing law is for this kind of dispute, the opinion put it plainly:

Little to no existing caselaw directly dealing with how to ascribe responsibility for AI agents like the Assistant.

The case returns to the Northern District of California for further proceedings; Amazon says it is weighing a rehearing request or a Supreme Court appeal.

What it means for operators

This is the first significant appellate read on how the CFAA treats an AI agent acting on a user’s behalf, and it lands in favor of agent-makers: user intent, not the tool executing it, is what triggers “access” under the statute. If you build or operate agents that navigate third-party sites on a user’s behalf, this ruling gives you a real legal argument against a CFAA claim, though the court was explicit that its holding is tied to this specific record and does not set a broad rule for agentic AI generally. It’s a data point, not a green light. Keep respecting terms of service and site-level blocking regardless of what this ruling does to the hacking-statute exposure, and it doesn’t resolve the separate contract and IP disputes playing out in cases like Runlayer’s lawsuit against Rippling over its MCP gateway or the safety lawsuit against xAI.