Amazon Blocks Meta Muse

In August, the Ninth Circuit vacated Amazon's injunction against Perplexity's AI shopping agent, ruling that it's the user, not the AI company, who accesses a website under federal anti-hacking law. I wrote at the time that enterprises shouldn't read that as a green light for agentic commerce. It was one data point in a legal framework still being written one case at a time.

Six weeks later, here's the proof. Amazon blocked Meta's new AI agent, Muse, from shopping on its site, and it didn't touch the anti-hacking argument the Ninth Circuit rejected. The message Muse users now see cites Amazon's Conditions of Use instead: continued access by an unauthorized AI agent violates the agreement customers accepted.

That's not a coincidence, it's the workaround the court left open. The ruling closed the door on Amazon using the Computer Fraud and Abuse Act against agents acting on a user's behalf. It never touched contract law. Amazon lost its petition for a rehearing on September 10, and the following weekend it pivoted straight to the avenue that ruling never closed.

The irony is that Amazon is also the company running Muse's compute. Meta signed a multibillion-dollar deal in April to power Muse's agentic workloads on Amazon's own Graviton chips. Amazon will sell Meta the engine and still block that same engine from its own store, using the one legal lever the courts left standing.

If you're building or deploying AI agents right now, the real lesson isn't the CFAA ruling itself. It's that platforms have more than one way to keep agents out, and they'll use whichever one the courts leave open.

What's the more durable fix here, new legislation or platforms and agent builders negotiating terms directly?