Amazon.com Services v. Perplexity AI – Ninth Circuit Vacates Preliminary Injunction

Bottom line

The U.S. Court of Appeals for the Ninth Circuit vacated the district court’s preliminary injunction against Perplexity AI’s Comet browser AI assistant and remanded for further proceedings, concluding that Amazon is unlikely to prove the assistant “accessed” its servers under the federal Computer Fraud and Abuse Act (CFA A) or California’s Comprehensive Computer Data Access and Fraud Act (CDAFA).

Case overview

Parties: Amazon.com Services, LLC (plaintiff) vs. Perplexity AI, Inc. (defendant)
Court: Ninth Circuit, decided August 4 2026 (Docket 26‑1444)
Issue: Whether Perplexity’s AI‑enabled browser tool, Comet, unlawfully accessed Amazon’s website in violation of the CFAA and CDAFA.
District court: Granted a preliminary injunction, finding Amazon likely to succeed on the merits and to suffer irreparable harm.
Appeal outcome: The Ninth Circuit vacated the injunction and remanded, holding that the statutory “access” requirement was not satisfied because the user, not Perplexity, accessed Amazon’s servers.

Why the Ninth Circuit vacated the injunction

1. Access requirement not met

The CFAA requires a defendant to intentionally access a protected computer. 18 U.S.C. § 1030(a)(2).
The court emphasized that access means “entering” a computer system, not merely sending instructions to a third‑party server.
Perplexity’s assistant never directly contacted Amazon’s servers; it only received screenshots from the user’s browser and sent them to Perplexity’s own servers for processing.
The court therefore concluded that the user accessed Amazon’s site, with the assistant acting as a tool.

2. Same reasoning applies to the CDAFA

California’s CDAFA defines “access” similarly and also requires a person to cause the unauthorized access.
Because the assistant is a software tool, the court found the user, not Perplexity, was the actor who accessed Amazon.

3. Equitable factors favor Perplexity

Irreparable harm: Amazon’s evidence of harm (potential degradation of shopping experience, speculative cyber‑risk) was deemed weak.
Balance of equities: An injunction would unduly burden Perplexity’s development of a novel AI product while providing little benefit to Amazon.
Public interest: Limiting agentic AI would stifle innovation and consumer choice, contrary to public policy.

Key excerpts from the opinion

"It is the user who ‘accesses’ Amazon’s computers, with the help of the Assistant to carry out specific acts on Amazon.com. … Perplexity may receive screenshots and send instructions, but those activities do not mean that Perplexity has ‘accessed’ (gained entry) to Amazon’s servers."

"Because we recognize that agentic AI is an emerging technology, we do not establish a new legal regime governing agentic AI. Our holding is limited to ‘access’ as contemplated by the CFAA and as applied to the Assistant’s interactions with Amazon.com on the record before us."

Reactions from the Hacker News community

Commenter	Insight	Takeaway
gz5	Frames the dispute as a business threat: headless AI browsing could undercut Amazon’s ad revenue.	Highlights the economic stakes beyond the legal question.
eigencoder	Argues Amazon lacks standing because browsers already act on users’ behalf.	Mirrors the court’s view that the user, not the tool maker, is the actor.
theturtletalks	Warns that LLM‑driven agents could become a new marketplace layer, potentially displacing platforms like Amazon.	Suggests broader industry disruption if agentic AI proliferates.
cmiles8 (first comment)	Notes the appellate reversal flips liability onto users, implying Amazon must target customers rather than the AI provider.	Reinforces the court’s attribution of conduct to the user.
dzonga	Calls the ruling a welcome clarification for search‑engine‑type services that scrape data.	Views the decision as a win for data‑aggregation businesses.
keeda	Points out that anything a human can legally do can be delegated to software, provided no security measures are bypassed.	Aligns with the court’s emphasis on user‑directed actions.
binlog	Observes Amazon isn’t suing OpenAI or Anthropic, likely because they run on AWS, hinting at strategic considerations.	Suggests selective enforcement based on business relationships.
Legend2440	Provides a concise TL;DR: Amazon got a preliminary injunction, the Ninth Circuit vacated it, and the case is still pending on the merits.	Summarizes the procedural posture for readers.

Implications for AI‑enabled browsing tools

Tool makers are not automatically liable for CFAA/CDAFA violations when their software acts on a user’s instructions, unless the tool itself directly contacts the target server.
User‑agent strings (or lack thereof) are not determinative; the statutory analysis focuses on who accesses the protected computer.
Future litigation will likely hinge on whether an AI system can be said to independently initiate communication with a target server, rather than merely relaying user‑driven actions.
Business models that rely on AI agents to automate transactions may need to negotiate explicit API agreements or licensing terms with platforms to avoid contractual disputes, even if criminal statutes are inapplicable.

What remains unsettled

The Ninth Circuit remanded for further proceedings; the district court must now address the merits without the preliminary injunction.
The court did not address tort claims, contract claims, or privacy implications of the assistant’s screenshot collection.
As agentic AI evolves, courts will eventually need to decide whether more autonomous agents cross the line from “user‑directed tool” to “independent actor” under the CFAA.

Bottom line for developers and businesses

Design AI assistants so that all communication with third‑party servers is mediated by the user’s browser, not by the AI provider’s servers.
Include clear user‑agent identifiers and respect site‑specific robots.txt or API terms to avoid contractual disputes.
Anticipate that legal risk will focus on direct server access; indirect data collection (screenshots, logs) is less likely to trigger CFAA liability.
Monitor future case law for shifts in how courts treat increasingly autonomous AI agents.