A Federal Court Just Ruled Your AI Agent Is Legally You — Not the Company Behind It

The most important question in AI law right now is deceptively simple: when an AI agent visits a website on your behalf, who is doing the visiting? You, or the company that built the agent? On August 4, 2026, the Ninth Circuit gave its answer — and it used an analogy so clean that it may define this area of law for years. The answer matters not just for Perplexity and Amazon, but for every company building AI that acts in the world on a human’s instructions.

What exactly did the court decide?

The U.S. Court of Appeals for the Ninth Circuit vacated a preliminary injunction that had barred Perplexity’s AI agent — the “Assistant” feature in its Comet browser — from accessing Amazon.com on customers’ behalf.

The background: Amazon had sued Perplexity, alleging that its agentic browser feature accessed users’ password-protected Amazon accounts to browse and purchase products at users’ direction, without identifying itself to Amazon as an AI agent, in violation of Amazon’s terms of service. Amazon brought claims under the Computer Fraud and Abuse Act (CFAA) — the federal computer fraud statute — and California’s equivalent, the Comprehensive Computer Data Access and Fraud Act (CDAFA).

A lower court had issued an injunction blocking Perplexity’s agent from accessing Amazon. The Ninth Circuit reversed, holding that Amazon was unlikely to succeed on its CFAA and CDAFA claims against Perplexity.

The reasoning is what makes this ruling genuinely significant.

That single analogy resolves a question that has been hypothetical in legal scholarship for years and is now binding circuit court precedent. This is the first circuit court ruling to address the intersection of agentic AI and the CFAA — a genuinely significant legal first.

Why does the “browser analogy” matter so much?

Because it draws the legal identity line at the human, not the tool. And because it does so using logic that is both intuitive and difficult to limit to a single case.

Consider what the alternative ruling would have meant. If the court had held that Perplexity — the company — was the one “accessing” Amazon when its agent acted on a user’s instructions, then every AI agent that interacts with a third-party website would be legally the AI company accessing that site. Every agentic action would be the company’s action, subject to every website’s terms of service, potentially violating computer fraud statutes with every click.

That would have made agentic AI — AI that does things in the world on your behalf — functionally illegal as a product category, because no AI company could guarantee compliance with every website’s terms on every action a user might instruct.

The court’s reasoning avoids that outcome with a principle that is both narrower and more durable: the human who gives the instruction is the one who acts. The AI is the tool, like a browser is a tool. The legal responsibility follows the human.

What does this mean for the broader AI browser landscape?

The ruling lands at a moment when the agentic browser market is in visible flux.

Perplexity’s Comet is expanding. The Comet browser — whose AI agent feature was at the center of this case — has reportedly been growing its reach, including to Samsung Internet and enterprise deployment via MDM (mobile device management) for organizations. The Ninth Circuit ruling removes the legal cloud that hung over its core feature and gives Perplexity a first-circuit-court endorsement of the principle that its agent’s actions are legally its users’ actions.

OpenAI’s ChatGPT Atlas, meanwhile, reportedly went the other direction. Around August 9, 2026, Atlas — OpenAI’s standalone agentic browser, launched roughly nine months earlier in October 2025 — reportedly stopped working as a standalone product. Its agentic browsing features were reportedly folded into the ChatGPT desktop app and a Chrome extension instead. The shift from standalone browser to embedded features inside an existing app represents a different product bet: rather than owning the browser, own the layer that sits on top of whatever browser the user already has.

What questions does this ruling leave open?

Several significant ones, because the court answered a specific question — who is “accessing” a website under the CFAA — and deliberately did not answer the broader questions around it.

Terms of service enforcement is untouched. The ruling says the user is the one accessing the site, which means CFAA claims against the AI company are unlikely to succeed. But Amazon’s terms of service still prohibit automated access. The question of whether Amazon can enforce those terms against individual users who use AI agents — or against Perplexity on some other legal theory — is not resolved by this ruling.

Disclosure obligations are unaddressed. Amazon’s complaint alleged that Perplexity’s agent accessed user accounts “without identifying itself as an AI agent.” The court’s browser analogy sidesteps this: Safari does not identify itself as “Apple’s browser” when it loads Amazon, so the lack of AI-agent identification is not the CFAA violation Amazon claimed. But whether there should be a disclosure obligation — separate from the CFAA — is a policy question this ruling does not reach.

Scaling effects are unexplored. One user tasking one AI agent to buy one product on Amazon is legally clean under this ruling. But what about a thousand users whose agents all execute coordinated purchasing strategies? Or an AI agent that monitors prices across a retailer’s entire catalog continuously? The browser analogy holds at the individual level, but its limits at scale are untested.

Question	Status after the ruling
Who “accesses” a website when an AI agent acts for a user?	Answered: the user
Can Amazon enforce TOS against AI agent users on other legal theories?	Open
Must AI agents identify themselves to websites?	Unaddressed
Does the browser analogy hold at scale (coordinated actions, continuous monitoring)?	Untested
Can other circuits reach different conclusions?	Yes — this is Ninth Circuit only

Why this is the most important AI ruling of 2026 so far

Because it answers the foundational question that every agentic AI product depends on, and it answers it in favor of the user, using reasoning that is intuitive, precedent-compatible, and difficult to cabin to a single fact pattern.

Every AI company building agents that interact with the web — shopping, booking, form-filling, data gathering, anything that touches a third-party website — needed to know whether their product would be treated as the company accessing those sites (legally dangerous) or the user accessing those sites through a tool (legally familiar). The Ninth Circuit said it is the second thing.

The bottom line

A federal circuit court, for the first time, drew a line on the most fundamental question in agentic AI law: when your AI acts for you, it is legally you acting, not the AI company. The reasoning is a browser analogy so intuitive it may prove difficult to dislodge — and it hands the entire agentic AI industry the legal foundation it needed to build on. The questions it leaves open — around terms of service, disclosure, and scale — are the fights that come next. But the bedrock question is answered, and the answer is: you are the user, and your agent is your tool.