Quinn Emanuel wins Ninth Circuit reversal for Perplexity in landmark AI agent access case against Amazon
Aug 4, 2026 with John Quinn
Key Points
- The Ninth Circuit reversed a district court injunction blocking Perplexity's AI shopping agent Comet from Amazon's website, ruling that users—not Perplexity—access Amazon's systems and therefore no Computer Fraud and Abuse Act violation occurred.
- Amazon's legal challenge was a vehicle for commercial objection: Comet generates no ad revenue and cannot be upsold, exposing a gap between traditional computer-access law and AI agent operations.
- Courts will now map agency law, contract formation, and tort liability onto AI agents that can execute thousands of transactions autonomously, with standard-form contracts and technical safeguards likely to emerge as industry norms.
Summary
Read full transcript →Perplexity wins Ninth Circuit reversal against Amazon
The Ninth Circuit Court of Appeals reversed a district court ruling that had barred Perplexity's AI shopping agent, Comet, from operating on Amazon's website. Quinn Emanuel, representing Perplexity, received the opinion within hours of the conversation — the general press had not yet reported it.
Amazon sued Perplexity under the Computer Fraud and Abuse Act, which requires that a defendant access a computer without permission, extract information, and cause damage. Amazon's argument was that Comet violated those terms by entering its website without consent. The district court in San Francisco agreed and granted Amazon's preliminary injunction. The Ninth Circuit reversed on a straightforward point: it is the user, not Perplexity, who accesses Amazon's systems. The user downloads Comet, deploys it, and directs it to shop. Perplexity never touches Amazon's computers directly.
John Quinn notes that Amazon's real objection is commercial — Comet has no eyeballs, sees no ads, and cannot be upsold. The legal argument was a vehicle for a business interest.
“Amazon sued Perplexity under the Federal Computer Fraud and Abuse Act. Amazon's suing Perplexity saying, your agent is being deployed. We don't want that to happen. Very unfortunately, the district court up in San Francisco entered a preliminary injunction a few months ago. But just two hours ago, we got the opinion from the Ninth Circuit. The Ninth Circuit understood that Perplexity is not accessing Amazon's computer — it's the user.”
What the ruling opens up
Quinn frames the decision as an early data point in a much larger unsettled area. Traditional legal concepts — authority, apparent authority, contract formation, tort liability — all have to be mapped onto a world where AI agents can enter into thousands of contracts before anyone notices something has gone wrong.
The contract risk is concrete. An agent instructed to buy compute at the cheapest price might lock the principal into a long-term deal with a heavy cancellation penalty. California Civil Code Section 1714, enacted this year, explicitly bars blaming the agent — the principal is responsible. Courts will likely apply familiar agency-law concepts: did the agent have actual authority, and if not, did it have apparent authority sufficient to bind the principal? Quinn expects standard-form contracts designed specifically for agent transactions to emerge, alongside technical safeguards — spending caps, human-in-the-loop requirements for material transactions, and systematic auditing of transaction histories.
Tort exposure is also live. Quinn points to cases already being filed alleging that AI agents persuaded users toward self-harm, raising questions about whether software qualifies as a defective product under products liability doctrine. He also flags trespass to chattels — a traditional claim covering unauthorized access to personal property, including servers — as a theory that has been raised but not yet gained significant traction.
Copyright and the pace of precedent
On training data, Quinn says early signals from a handful of cases suggest that training on copyrighted material may qualify as fair use, but nothing is settled. The full body of answers will come through the courts case by case, some reaching the Ninth Circuit and potentially the Supreme Court. Legislation is possible but slow — and until Congress acts, judges will build the framework one opinion at a time.
The billing model problem for law firms
Most large law firms bill by the hour. Quinn's candid read is that the profession is largely "whistling in the dark" about what AI will do to that model. If a lawyer can produce a work product that is 80 to 90 percent complete at the push of a button, the hours simply vanish — and clients will expect the savings to flow through. Big Law is currently profitable and has little short-term incentive to restructure. The AI-related litigation surge, including novel disputes like the Perplexity case and the Altman-Musk dispute, is masking the structural pressure for now.
Data center contracts
Quinn expects some data center deals to end in reorganization. The financial structures are complex — SPVs, bond financing, securitizations, private credit, cross-default provisions — and not every project will come online on schedule. Supply constraints are real: one neo-cloud told Quinn the biggest bottleneck is electricians. When deals break down, liability management exercises, workouts, and bankruptcies follow, and lawyers will be deeply involved in all of it.
Every deal, every interview. 5 minutes.
TBPN Digest delivers summaries of the latest fundraises, interviews and tech news from TBPN, every weekday.