IP Law Daily, TECHNOLOGY/INTERNET—9th Cir.: Amazon’s unlawful access claims against Perplexity AI unlikely to succeed; injunction vacated, (Aug 7, 2026)

By Glenn Borst, J.D., LL.M.

Individual users of an AI-powered browser assistant, not the assistant itself or the company that created it, accessed Amazon’s servers.

In a widely monitored dispute addressing the use of agentic artificial intelligence (AI) in online commerce, the U.S. Court of Appeals for the Ninth Circuit has held that an AI-powered browser assistant created by Perplexity AI did not “access” Amazon’s servers in violation of federal and California laws against computer hacking. Rather, the individual user of the assistant accessed Amazon’s servers. Therefore, Amazon’s suit against Perplexity was unlikely to succeed on the merits, and a preliminary injunction against Perplexity granted by the district court was vacated (Amazon.com Services, LLC v. Perplexity AI, Inc., No. 26-1444 (9th Cir. Aug. 4, 2026)).

Perplexity created an AI-enabled browser, called Comet, which includes an optional AI assistant that can assist the user in browsing websites, such as Amazon, to shop for requested merchandise. In its lawsuit, Amazon alleged that Perplexity, through its AI assistant, unlawfully accessed Amazon's systems in violation of the federal Computer Fraud and Abuse Act (CFAA) and California's Comprehensive Computer Data Access and Fraud Act (CDAFA). Amazon argued that Perplexity's software navigated Amazon's website and collected information from user accounts despite such activity being prohibited by Amazon. A federal district court agreed with Amazon and granted a preliminary injunction in March 2026 after concluding that Amazon was likely to succeed on its claims and that Perplexity’s actions had caused Amazon to incur significant costs responding to the activity. On appeal, the Ninth Circuit sided with Perplexity, vacating the lower court decision and the preliminary injunction and remanding for further proceedings.

How the AI assistant operates. When a Comet user directs the AI assistant to locate an item on Amazon’s website, the assistant takes screenshots of the browser view and sends them from the user’s computer to Perplexity’s servers. Those servers then instruct the assistant on how to navigate Amazon.com. The assistant does not operate independently and relies on direction from the user and instructions from Perplexity’s servers.

Specifically, when a user communicates with a website, only the user’s browser communicates with that website’s server. The browser requests information from the website and displays that information to the user. If the user activates the AI assistant, the assistant then analyzes the contents of the page as displayed on the user’s computer and then sends that information, along with the user’s instructions, to Perplexity’s AI servers. Perplexity’s servers never directly access the Amazon servers.

Preliminary injunction determination. The district court abused its discretion in granting a preliminary injunction to Amazon, according to the appellate court. The CFAA requires that Perplexity “intentionally accessed a computer” without authorization, obtaining information that resulted in loss to one or more persons of at least $5,000. However, no Perplexity computer gained access to Amazon’s servers, but rather, any Amazon data was transmitted to the user’s computer, and then from there to Perplexity’s servers. The assistant is computer software, and it has no “intent” other than what the user directs it to do on behalf of the user.

Emerging law on agentic AI. Amazon argued that the assistant’s agentic capabilities, that is, its autonomous actions, went beyond the passive displays of a traditional web browser. The appellate panel noted that agentic AI is an emerging technology, and that there is little case law on how to ascribe responsibility for AI agents like the assistant. The court has previously explained that because the CFAA is primarily a criminal statute and interpretations of its provisions are equally applicable in the civil and criminal contexts, the rule of lenity should guide any interpretation of the statute, meaning that any ambiguity should be construed against the government.

Who is the “user.” The appellate panel emphasized that the distinction between an AI tool and the user controlling it was critical under current computer-access laws. Perplexity's technology assisted users in navigating Amazon's website, but did not independently access Amazon's systems in the manner required to establish liability under the CFAA or CDAFA. In creating the CFAA, Congress aimed to prevent “intentional intrusion onto someone else’s computer--specifically, computer hacking.” The CFAA contemplates access by a “person,” and the AI assistant is tool, not a person, for statutory purposes. Only the user accesses Amazon’s servers. Amazon was therefore unlikely to succeed on the merits of its claim.

State law claim. The CDAFA is similar to the CFAA, although with a few differences. Amazon argued that the CDAFA had a broader definition of “access,” simpler permission requirements, and no requirement that information be “obtained.” However, the state law has the same applicability to “any person” who causes unauthorized access to a computer. Thus, Amazon was unlikely to succeed on this claim.

Other equitable factors. Even if a plaintiff is likely to succeed on the merits, the district court must nonetheless consider other factors in its analysis, namely, irreparable injury, the balance of the equities, and the public interest. Likelihood of success is not sufficient on its own to warrant a preliminary injunction. The appellate court found that, in contrast to cases where public goodwill toward a company is damaged, the degradation of the overall Amazon.com shopping experience was more abstract. Arguments that Perplexity posed a security risk to Amazon were also rejected as not demonstrated, and as mitigated by improvements made by Perplexity to its product. The balance of equities and the public interest also did not aid Amazon’s claim. Given the unlikelihood that Amazon would succeed on its claims, the injunction would impose an inequitable burden on Perplexity, impair consumer choice, and needlessly limit development of a new technology. These factors therefore weighed against an injunction.

Going forward. The case is not over, and continues after remand for further proceedings, where broader questions regarding AI-assisted browsing, website restrictions, and platform control may continue to be litigated. For tech companies working on the development of AI agents, this decision represents an early appellate endorsement of the view that AI assistants may be treated as user-directed tools rather than independent actors. For website operators such as Amazon, the ruling highlights the challenge of applying decades-old computer access laws to rapidly evolving AI technologies.

The Case is No. 26-1444.

Judge: Smith, M.

Attorneys: Hagan Scotten (Hueston Hennigan LLP) for Amazon.com Services, LLC. Christopher G. Michel (Quinn Emanuel Urquhart & Sullivan LLP) for Perplexity AI, Inc.

Companies: Amazon.com Services, LLC; Perplexity AI, Inc.

MainStory: TopStory AINews TechnologyInternet AlaskaNews ArizonaNews CaliforniaNews HawaiiNews IdahoNews MontanaNews NevadaNews OregonNews WashingtonNews