Overview
A federal judge in San Francisco has moved to curb a fast rising feature of generative AI: agents that shop and transact on behalf of users. On Monday, March 9, 2026, U.S. District Judge Maxine Chesney issued a preliminary injunction that prevents Perplexity AI from using its Comet browser agent to access password protected areas of Amazon, including Prime accounts. The ruling also orders Perplexity to destroy any Amazon data it has already collected and pauses for one week to allow an appeal. The original reporting on the case appears in Search Engine Land by Danny Goodwin. (See the full item on Search Engine Land and the related Bloomberg report.)
The ruling in brief
– The injunction bars Perplexity from using Comet to sign in to or otherwise access account pages on Amazon, on the grounds that the agent accessed accounts with users permission but without authorisation from Amazon.
– Judge Chesney found that Amazon presented strong evidence of unauthorised access and ordered destruction of collected Amazon data.
– The order is stayed for seven days to permit Perplexity to seek immediate appellate review. If Amazon’s motion survives, AI developers will face immediate practical limits when agents try to operate inside logged in sessions on major e commerce platforms.
Why this matters to business and consumers
The case tests a central promise of autonomous agents: that they can carry out complex, credentialed tasks on a user’s behalf, from reordering household staples to completing checkout. If courts require explicit website authorisation for agents to operate inside personal accounts, the business model for many assistant features will become far more complicated.
For platforms such as Amazon, the ruling reinforces a simple proposition: account access is part of a trusted commerce relationship and must be protected. For AI vendors, it means product teams will need to design safer, more transparent integrations that either rely on authorised APIs or secure tokenised consent flows. Perplexity’s Comet reportedly made purchases while not clearly identifying itself as a bot, and that lack of transparency sits at the heart of Amazon’s claims. See Search Engine Land for the original coverage by Danny Goodwin and the Bloomberg account of Amazon’s statement.
Legal and technical fault lines
Amazon sued Perplexity in November on grounds of computer fraud and unauthorised access. The preliminary injunction suggests courts are willing to grant fast relief when a large platform claims a consumer trust or security injury. Practically, that may push AI firms to adopt one of two routes:
– Build formal integrations and partnerships with major sites using documented APIs and mutual authentication.
– Limit agents to non authenticated interactions, handing back to users the final click or payment step.
Either route raises costs. Startups that previously relied on agent behaviours that mimic human browsing will now have to negotiate contracts or rebuild their stacks. That shift benefits incumbents who can offer authorised, paid integrations but could stifle the rapid experimentation that has driven AI innovation.
A wider ripple effect
Although this is a US federal case, the commercial stakes are global. E commerce platforms, payment providers and banks will watch closely. Companies operating in or from Ireland and the wider EU should note how US litigation over agent behaviour could influence future negotiations around data access, user consent and platform APIs. Regulators and courts in Europe are increasingly attentive to automated access and consumer protection, and a string of American rulings could accelerate commercial pushback against unauthorised automation.
Market and policy implications
From a policy perspective, the ruling emphasises the need for clear industry standards for agent behaviour. That includes technical labels for automated clients, standardised authentication flows and audit logs showing a user explicitly authorised a bot to act in their account. For businesses, the immediate lesson is risk management: allow only authorised or vetted agent activity on platforms where money or personal data is at stake.
Reaction and next steps
Amazon described the injunction as necessary to protect customers and the shopping experience. Perplexity has one week to appeal the preliminary order. If the appeal is filed, the legal fight will proceed faster than a typical discovery heavy case, and the technology community will be watching the interplay between court orders, platform policies and product roadmaps.
Readers who want the primary reporting can read Danny Goodwin’s piece on Search Engine Land and the Bloomberg update. For Perplexity’s perspective see the company website. Amazon’s public statements around platform security and user trust are available on its corporate site.
What businesses should do now
– Review any agent or automation that interacts with third party accounts and map where credentialed access occurs.
– Prioritise formal integrations and avoid mimicking human sign in flows when a documented API or partnership is available.
– Prepare disclosure and consent mechanisms that are auditable and explicit so that user authorisations are clear in any future dispute.
The court’s move is a reminder that technological possibility does not guarantee legal or commercial acceptance. For firms building assistants and autonomous agents the path forward will be technical, contractual and legal. The coming weeks will show whether Perplexity can regain the capability in question through appeal or whether the industry will steer toward authorised, standardised integrations.
