Law & Regulation

Your Site Cannot Use A Hacking Law Against An AI Agent

A US appeals court ruled on 4 August that Amazon cannot use the federal hacking law against Perplexity's AI shopping agent, because the user, not the tool, accesses the site. European operators should not assume their own computer-misuse laws would rule differently.

Law & RegulationBy Servola Tech Desk2026-08-054 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

Your Site Cannot Use A Hacking Law Against An AI Agent

Key takeaways

On 4 August, the Ninth Circuit vacated a preliminary injunction that had barred Perplexity's Comet browser from acting as a shopping agent on Amazon, ruling that Amazon is unlikely to win its core legal claim because the user, not Perplexity, is the one who accesses Amazon's site under the federal Computer Fraud and Abuse Act.
The panel's reasoning is narrow but consequential: an AI agent is a tool, not a person for statutory purposes, so a law written to punish unauthorized human access to a computer does not reach software a user chooses to act on their behalf, at least not on this record.
This is not the end of the case. The court only lifted the injunction pending further proceedings and left Amazon free to pursue other claims, such as breach of its terms of service, which do not require proving unauthorized computer access.
For a European operator, the transferable lesson is the mechanism, not the jurisdiction: contract enforcement, not a hacking statute, is emerging as the reliable lever against an AI agent a platform does not want visiting its site, whether the underlying law is the US CFAA or the UK's Computer Misuse Act.
Any business publishing a website that AI agents can now browse, fill forms on, or transact through should audit its terms of service for an explicit no-agent or no-automated-access clause now, because a contract claim, not a computer-crime claim, is what a court is likely to entertain next.

What The Ninth Circuit Actually Decided On 4 August

The case began on 9 March 2026, when a district court granted Amazon a preliminary injunction stopping Perplexity's Comet, an agentic web browser, from using its automation to navigate password-protected parts of Amazon's site on a user's behalf. Amazon argued the tool was violating the federal Computer Fraud and Abuse Act (CFAA) and its California state-law equivalent by accessing Amazon's systems without authorization. Perplexity appealed, the Ninth Circuit stayed the injunction pending the outcome, and on 4 August a three-judge panel ruled: Amazon is unlikely to succeed on the CFAA claim, because the statute punishes unauthorized access by a person, and here it is the user, logged into their own Amazon account, who is doing the accessing. The court described Perplexity's software as a tool the user directs, not an independent legal actor.

The Electronic Frontier Foundation, which filed an amicus brief backing Perplexity, framed the stakes plainly: accepting Amazon's theory would mean the browser you use, the extension you install, or the assistant you delegate a task to could itself commit a federal crime every time it fetches a page on your behalf, exposing you, the user, to liability by extension. The panel did not have to accept that framing to rule for Perplexity, but its holding, that software is a tool and not a person for CFAA purposes, forecloses the version of that argument that treats an agent's actions as a discrete unauthorized access separate from the user who launched it.

Why The Ruling Is Narrower Than The Headline Suggests

Vacating a preliminary injunction is not a final judgment. The Ninth Circuit only decided Amazon was unlikely to win on the merits of its CFAA claim, which is the standard needed to justify blocking Perplexity before trial; it did not dismiss the case, and it left the door open for Amazon to develop other theories at the district court, most obviously a breach-of-contract claim under Amazon's own Conditions of Use, which every account holder has already agreed to and which does not require proving unauthorized computer access at all, only that the terms were broken.

That distinction is the actual lesson. A computer-misuse statute is written to punish intrusion, and courts are visibly reluctant to stretch that language to cover software a legitimate, logged-in user chose to run. A contract, by contrast, can say whatever the platform wants it to say, including an explicit ban on automated agents, and a court does not need to reinterpret a criminal-hacking law to enforce it. Any platform that wants to keep AI agents out is now watching two different legal tools produce two very different outcomes.

The Question Every European Site Operator Should Ask Now

The CFAA is a US statute, but the underlying question is not American: does a computer-misuse law, written before agentic software existed, treat an AI tool acting on a user's instruction as a form of unauthorized access? The UK's Computer Misuse Act 1990 and various EU member states' own cybercrime statutes use similarly broad, human-intrusion-era language, and none of them have yet been tested against an agentic browser the way the Ninth Circuit just tested the CFAA. A European operator cannot assume a local court would read its own statute the way a US judge just declined to.

The practical move is not to wait for that test case. Review the terms of service on any site that handles logins, purchases, or account data, and add explicit language on automated and AI-agent access now, because that clause, not the hope that a computer-crime law will do the work, is what determined the outcome of this case once the criminal-statute argument failed. A platform's real defense against an unwanted agent is the contract it wrote, not the hacking law it hoped would apply.

Frequently asked questions

What did the Ninth Circuit actually rule on 4 August 2026?

It vacated a preliminary injunction that had stopped Perplexity's Comet agentic browser from acting on Amazon's site on a user's behalf, ruling that Amazon is unlikely to win its claim under the federal Computer Fraud and Abuse Act because the user, not Perplexity, is the one accessing Amazon's systems. The case itself continues; only the injunction was lifted.

Does this mean AI shopping agents are now legal everywhere?

No. The ruling addresses one US federal statute and one specific injunction. Amazon can still pursue a breach-of-contract claim under its terms of service, and courts in other jurisdictions, including the UK and EU member states, have not ruled on how their own computer-misuse laws would treat the same conduct.

What should a European business with a website do about this?

Review your terms of service now for explicit language covering automated tools and AI agents, rather than assuming a hacking or computer-misuse statute will block unwanted agent traffic. This ruling shows a contract claim is currently the more reliable legal lever, and that lesson does not depend on which country's law applies.

A federal appeals court just settled who is legally driving when an AI agent visits a website: the user, not the software. Every platform that wants a say in who visits should stop assuming a hacking law will make that call for them.

PerplexityAmazonNinth CircuitCFAAAI agentsComputer Misuse ActTerms of service

Nothing Crashed, and It Cost $1.8 Million

An Amazon AI job ran five months, cost 1.8 million dollars and never shipped. Two months earlier the company killed a leaderboard for the same underlying reason.

5 min read

Capital

AWS Now Earns 60 Percent Of Amazon's Profit

Amazon reported 27.5 billion dollars of operating income for Q2 2026 and AWS supplied 16.6 billion of it. The guidance for next quarter is lower. Here is why that matters at renewal.

4 min read

Law & Regulation

The EA Buyout Now Hinges on Brussels

The largest take-private in history has cleared EA shareholders and US antitrust. The last real gate is the EU - and not classic merger review, but the Foreign Subsidies Regulation, facing its first blockbuster test.

3 min read

← All articles