The editorial frames the pivotal question as whether a website's ToS can transform a user's own delegated automation into a federal crime, and invokes hiQ v. LinkedIn and Van Buren as precedent narrowing the CFAA to require bypassing an access control rather than merely violating a policy. Since the user is logged in with their own credentials and lending that authorization to software, the editorial implies this shouldn't rise to unauthorized access.
Amazon argues its Terms of Service prohibit automated access, and that Perplexity — by shipping Comet as a tool that facilitates automated checkout at scale — is trafficking in unauthorized entry to Amazon's computers. The underlying commercial stake is Amazon's ~$50B advertising business, which agent-mediated shopping bypasses entirely.
By surfacing the Ninth Circuit filing to the HN front page (203 points, 201 comments), the submitter signals that the developer community sees this as a landmark case whose outcome will define the legal boundaries for every agentic browser product going forward, not just a one-off dispute between two companies.
The editorial opens by predicting this case 'will probably be cited in every agentic-browser lawsuit for the next decade,' positioning it as the foundational precedent that will determine whether user-delegated automation is legally viable as a product category.
On August 4, 2026, the U.S. Court of Appeals for the Ninth Circuit took up *Amazon.com Services, LLC v. Perplexity AI, Inc.* — a case that will probably be cited in every agentic-browser lawsuit for the next decade. Amazon's complaint alleges that Perplexity's Comet browser, when instructed by a logged-in user to buy something on Amazon.com, is committing unauthorized access under the federal Computer Fraud and Abuse Act (CFAA) and California's Comprehensive Computer Data Access and Fraud Act (CDAFA).
The underlying facts are boring in a way that makes the legal question sharp. Comet is a browser. A user opens it, signs into their own Amazon account with their own credentials, and asks the agent to buy a specific product. Comet loads amazon.com, navigates the DOM, clicks buttons, and completes the checkout. Nothing is scraped anonymously; nothing is resold; there is no fake account. The user's cookies are the user's cookies. Amazon's position is that its Terms of Service prohibit automated access, and Perplexity — by shipping a tool that facilitates that access at scale — is trafficking in unauthorized entry to Amazon's computers.
The pivotal question is whether a website's Terms of Service can transform a user's own delegated automation into a federal computer crime. The Ninth Circuit has been here before, in *hiQ v. LinkedIn* and *Van Buren v. United States*, both of which trimmed the CFAA back toward a narrower reading: you need to *bypass* an access control, not just violate a policy. But those cases were about scraping public data. This one is about a logged-in session — the user has authorization, and the fight is over whether they can lend that authorization to a piece of software.
Start with the money, because that's what this case is actually about. Amazon's advertising business generated roughly $50 billion in 2024 and is one of the fastest-growing lines on its P&L. Sponsored product placements, brand ads, and the ranked search results page are only monetizable if a human eyeball lands on the page. A Comet user who says 'buy me a good USB-C hub under $30' never sees the ad slot, never scrolls past the sponsored results, and never bids up the auction that pays Amazon's rent. As one HN commenter put it, "headless Amazon makes it more difficult for Amazon to sell ads, which is a large part of their revenue." That is not a legal argument, but it is the argument.
The threat is structural. If agentic commerce becomes the default interface — you talk to Claude or ChatGPT, it talks to Amazon — the marketplace collapses into a wholesale API and Amazon becomes a fulfillment layer competing on price and delivery time against Walmart, Target, and Shopify's aggregate long tail. The brand equity of amazon.com the destination evaporates. The ad auction dies. The recommendation engine becomes irrelevant. This is why Amazon is fighting it in court instead of just adding a CAPTCHA — a CAPTCHA is an arms race Amazon loses, but a favorable CFAA ruling is a moat.
Perplexity's defense is philosophically simple: the user is the principal, the agent is the instrument, and the site cannot revoke a user's own permission to use their own account. From Perplexity's side, Comet is a browser. So is Firefox. So is a Playwright script an engineer wrote in an afternoon. If Amazon wins, the line between "browser" and "unauthorized automation" becomes whatever Amazon says it is on any given Tuesday. Every accessibility tool that scripts a checkout for a disabled user; every price-tracking extension; every corporate procurement bot; every family-account shared shopping list — all inherit a CFAA-shaped cloud.
The smart-people-disagree part is real. Amazon has a legitimate interest in defending against genuinely abusive automation — inventory hoarding by scalper bots, coupon fraud, review manipulation. The CFAA is a blunt tool for that job, but it is the tool that exists. And there is precedent for treating ToS-violating access as unauthorized in narrow circumstances, particularly when the operator has been sent a cease-and-desist and continues anyway. Perplexity, per the filings, was told to stop and did not.
But the counterweight is *Van Buren*, where the Supreme Court held that CFAA "exceeds authorized access" applies to accessing files you're not allowed to access, not to accessing files you are allowed to access for a purpose the owner dislikes. A user logged into their own Amazon account is squarely inside the authorized-access zone. Extending CFAA to reach the *manner* of that access — human fingers good, agent scripts bad — is the kind of expansion Van Buren explicitly rejected. That is the strongest ground Perplexity has, and it is strong.
If you're building anything that automates a logged-in third-party session, this case is now on your risk register. Assume every large marketplace and SaaS vendor is watching the Ninth Circuit and drafting their own cease-and-desist templates. The practical implications:
Ship with a documented user-consent trail. The strongest defense for any agent framework is that the user, not the vendor, is the entity taking the action. Log the prompt, log the account, log the intent. If you're building on top of Comet, Operator, browser-use, or your own Playwright rig, treat the audit log as a legal artifact, not a debug convenience.
Watch your ToS scrapers. If you have a compliance team, get an updated read on the ToS of every site your product touches with credentials. The old advice — "ToS violations aren't crimes" — is exactly what this case may narrow. Contracts of adhesion still bind you civilly even if they don't reach CFAA, and civil claims for tortious interference are already how sites like LinkedIn and Ticketmaster shut down automation vendors.
Design for the API future, not the DOM present. The stable equilibrium here is probably not "agents scrape retailers forever"; it's "retailers expose a commerce API and charge for access." Shopify already does this. Amazon's SP-API exists but is closed to shopper-side agents. If Amazon loses in court, expect a paid agent API within 18 months, priced to preserve the ad auction. If Amazon wins, expect the same, but with a CFAA hammer to enforce it.
On-device beats cloud for legally gray automation. An agent that runs in the user's own browser on the user's own machine using the user's own cookies is a much harder legal target than a SaaS that proxies the same requests through a datacenter. Comet's architecture — a real browser on the user's device — is not an accident.
The Ninth Circuit will probably rule narrowly and leave the interesting question for another case. But the strategic outcome is already visible: the era of building consumer agents on top of unwilling counterparties is ending, and the next generation will be either (a) licensed via paid commerce APIs, or (b) fully client-side and legally deniable. Anthropic's Computer Use, OpenAI's Operator, and every wrapper around them are converging on the same regulatory reality — the sites they operate against get a vote, and that vote will be cast in courtrooms until the economics of an agent-mediated internet get settled at the protocol layer. If you're building in this space, don't wait for the ruling. Assume the harder version of the outcome and design for it now.
I'm naive on the law around this, but it seems like Amazon.com shouldn't have standing here. What Perplexity does, from my perspective, is essentially the same as when I allow Firefox, Chrome, or Safari (or any other browser software) to see my credentials and access Amazon's website
I remember when I (naively, but not alone) thought that the Personal Computer would let us--consumers, individuals--express our own agency and priorities and control, a kind of democratized capital mini-factory anyone (or group) could tailor to their own needs.Nowadays it feels more like "Visit
> Amazon.com Services, LLC filed suit against Perplexity AI, Inc., an artificial intelligence company, asserting that Perplexity’s web browser tool, Comet, unlawfully accessed Amazon’s website in violation of the federal Computer Fraud and Abuse Act (CFAA) and California’s Comprehensive Computer
People don’t realize how big of a threat LLMs are to marketplaces like Amazon, because down the line people will just be talking to AI agents to help them find products, check out, do all of these things.So you’re probably like, why can’t I just use ChatGPT to do that? And you can, but ChatGPT is tr
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
Can't comment on the legal basis in the eyes of CFAA or DAFA, but from a business perspective AI is a legit threat to Amazon because headless Amazon makes it more difficult for Amazon to sell ads, which is a large part of their revenue.Meaning, even if merchants would have a difficult time movi