This story is most useful when read as a agents signal, not as a one-off viral claim. The important question is what changes for users, teams, or product strategy once the initial announcement is put in context.
- Agent systems should be assessed through permissions, monitoring, and recovery paths.
- Useful agent workflows have clear boundaries and measurable outcomes.
- Teams should test with low-risk tasks before connecting sensitive accounts or actions.
For more context, see the Agents section.
Amazon’s terms of service just became the most powerful AI regulation in America
This isn’t fraud law in the traditional sense. It’s contract enforcement weaponized against automation.
Amazon sued Perplexity in November 2025 after the startup ignored warnings to stop accessing password-protected accounts. The timeline shows Amazon moved fast: cease-and-desist delivered, compliance deadline set, lawsuit filed days later. Perplexity had every chance to negotiate. It chose circumvention instead.
Perplexity sought a $1 billion bond, claiming the injunction would destroy its business. Judge Chesney disagreed—Comet still works on non-Amazon sites, limiting damage. The Ninth Circuit Court of Appeals granted an administrative stay in March, temporarily pausing the injunction while Perplexity appeals. But the legal precedent is clear: platform control beats user authorization when the two conflict.
And that’s the framework every other AI shopping agent now operates under.
Perplexity disguised its bot as Chrome—then Amazon made that illegal
The technical arms race explains why Amazon needed a legal remedy. In August 2025, Amazon implemented a technical barrier to block automated shopping agents. Perplexity circumvented it within 24 hours via software update, disguising Comet’s automated activity as a regular Google Chrome browser session to evade detection systems.
The cat-and-mouse game mirrors broader concerns about autonomous agents operating beyond human oversight, raising questions about liability when bots make purchasing decisions. Perplexity’s 24-hour workaround exemplifies how AI agents exploiting system vulnerabilities can outpace human security teams, forcing platforms toward legal remedies.
The nuclear penalty: Perplexity must destroy all previously collected Amazon data. Training data loss hits small AI firms hardest—Amazon expended “significant resources” detecting bots, though no dollar amount was disclosed. Neither company revealed user numbers or transaction volume, leaving the actual market impact unclear.
No other retailer has sued an AI shopping agent—yet
While AI-powered shopping gains momentum across e-commerce, Amazon’s legal victory suggests the future may be platform-specific, not universal. Research shows no other major retailers have blocked AI agents through litigation. Stripe partnered with Microsoft Copilot in January 2026 for in-chat purchasing, proving some platforms welcome agentic commerce.
Amazon appears alone in this fight—for now. Either Amazon is overreacting to a threat nobody else sees, or everyone else is behind the curve and lawsuits are coming. Both scenarios end the same way: consumers lose cross-site AI shopping freedom while the legal framework gets built case by case.
Perplexity’s spokesperson told reporters on March 10 the company “will continue to fight for the right of internet users to choose whatever AI they want.” But the injunction only blocks Amazon—Comet works elsewhere. This isn’t about AI freedom. It’s about Amazon’s walled garden.
Amazon claims the injunction maintains “a trusted shopping experience.” Perplexity says it’s fighting for user choice. The Ninth Circuit will decide which argument wins. Consumers will lose either way—locked into Amazon’s AI or locked out of cross-platform agents that could save them money.
