Amazon vs Perplexity: Legal Battle Reshaping AI Commerce

Amazon’s legal victory over Perplexity settles nothing, but it signals clearly how incumbents intend to govern the agentic AI era entirely on their own terms.

Key Takeaways

• Judge Chesney’s preliminary injunction establishes that platform terms of service override individual user consent, setting a precedent that could fundamentally constrain how AI agents operate across commercial ecosystems.
• Amazon’s new Agent Policy, effective March 2026, signals that major platforms are moving from reactive litigation to proactive governance frameworks designed to control AI-driven commerce on their own terms.
• The dispute exposes a structural tension in agentic AI’s commercial model: startups like Perplexity depend on open platform access to deliver value, while incumbents have every financial incentive to restrict it.

A Courtroom Draws the Lines

U.S. District Judge Maxine Chesney granted Amazon a preliminary injunction against Perplexity AI, temporarily barring the startup’s Comet browser from accessing password-protected sections of Amazon’s platform. The order also mandates the destruction of any data extracted from Amazon’s systems and prohibits further unauthorized AI-driven purchases. It is a narrow ruling in legal scope, but its implications extend well beyond the immediate dispute.

What is being contested here is not merely a terms-of-service violation. It is a foundational question about the architecture of digital commerce in the age of autonomous AI: whether an AI agent, operating with a user’s explicit consent and credentials, can inherit that user’s rights to access a platform. Amazon says no. Perplexity says the opposite. The court, for now, has sided with Amazon, and the consequences for the broader industry deserve careful examination.

The Principals and Their Stakes

Amazon needs little introduction. Reporting $574.8 billion in net sales for fiscal year 2025, with AWS alone exceeding $100 billion in annual revenue, the company occupies a position of structural dominance across retail, cloud infrastructure, and increasingly, artificial intelligence. Its advertising business, which generated more than $50 billion in 2025, is central to understanding its objection to Comet. An AI browser that routes purchases while bypassing sponsored listings does not merely inconvenience Amazon; it directly undermines one of its most profitable and fastest-growing revenue lines.

Perplexity is a different kind of company entirely. Founded in 2022 as a conversational AI search engine, it has grown with unusual speed, raising over $1.5 billion in total capital and reaching a $20 billion valuation by September 2025, backed by investors including Nvidia and, notably, Jeff Bezos. Its Comet browser represents a strategic evolution from information retrieval toward task execution: an AI that does not merely answer questions about products but locates, compares, and purchases them autonomously. That proposition is commercially potent, and existentially threatening to platforms built around controlling the customer journey.

The Anatomy of the Dispute

Amazon’s complaint, filed in the U.S. District Court for the Northern District of California, centers on claims under the Computer Fraud and Abuse Act and California’s Comprehensive Computer Data Access and Fraud Act. The substance of the allegation is that Comet accessed Amazon customer accounts by mimicking human browser behavior, circumventing security measures and exposing the platform to fraud risk. Amazon documented at least five warnings issued to Perplexity since November 2024 and cited mitigation costs exceeding $5,000, a figure modest in isolation but legally sufficient to establish the standing required under federal statute.

Perplexity’s defense rests on the principle of delegated user authority. Users who deploy Comet do so voluntarily, providing their own credentials. The agent acts on their behalf, with their knowledge. In a published statement, the company argued that users have the right to choose their own AI tools and that platforms cannot unilaterally strip those rights simply because the tools in question reduce ad exposure or streamline the purchase funnel.

Judge Chesney found this argument unpersuasive. Citing the Ninth Circuit’s 2016 decision in Facebook v. Power Ventures, she held that platform revocation of access supersedes any consent granted by individual users. The user, in other words, cannot authorize what the platform has prohibited. That logic is clean in legal terms, and its implications for the agentic AI sector are immediate.

What the Ruling Actually Means

The injunction is preliminary, not final, and Perplexity has vowed to appeal. The legal contest will continue. But the ruling’s significance lies less in its immediate operational impact than in the framework it reinforces: platforms retain sovereign authority over their own systems, and AI agents that fail to operate within that authority do so at legal peril.

For AI startups building agentic products, this creates a structural dependency on negotiated access. To function reliably within major commercial platforms, agents will likely need formal integration agreements, not the informal assumption that user credentials constitute sufficient permission. That shifts leverage decisively toward incumbents. Amazon, Google, and their peers hold the keys; startups must come to the table or face exclusion.

This dynamic is already visible in Amazon’s operational response. The company updated its Business Solutions Agreement effective March 4, 2026, introducing an explicit Agent Policy requiring AI tools to self-identify, comply with platform rules, cease operations on demand, and refrain from using Amazon data to train third-party AI models. This is not litigation; it is governance. Amazon is not merely defending its platform after the fact. It is writing the rules prospectively, on its own terms.

The Competitive Contradiction

Amazon’s position carries a tension that Perplexity has been quick to note. In his Q3 2025 earnings remarks, CEO Andy Jassy spoke with evident enthusiasm about the potential of agentic commerce. Amazon has developed its own AI shopping tool, Rufus, which scans product data across the web and assists customers in navigating purchase decisions. The company is not opposed to AI-driven commerce; it is opposed to AI-driven commerce that it does not control.

That distinction matters. Over 180 small businesses filed protests in January 2026 against Amazon’s “Buy for Me” feature, which uses AI to facilitate purchases from third-party sites, a practice that critics characterize as applying to others precisely the kind of scraping behavior Amazon is litigating against Perplexity. Whether or not that characterization survives scrutiny, it points to the genuine complexity of establishing neutral, coherent rules in a market where the dominant platform is also a competitor.

The Regulatory Gap

The legal instruments deployed in this dispute, primarily the Computer Fraud and Abuse Act, were designed in an era when unauthorized computer access meant something considerably more straightforward than an AI agent placing a grocery order. The CFAA was written to address intrusion; its application to agentic commerce involves interpretive stretching that courts have managed with varying degrees of consistency.

For policymakers, this case illustrates the inadequacy of existing frameworks to address the specific competitive dynamics of platform AI. The question of whether user delegation can authorize agent access is not a hacking question; it is a commerce and competition question, and it may ultimately belong in legislation rather than litigation. Senior regulators in Washington and Brussels have the jurisdictional interest and the precedent from prior platform disputes to engage here, though the pace of regulatory response has historically lagged the pace of technological change.

An Industry at an Inflection Point

Perplexity’s broader ambitions remain intact despite the setback. The company launched Perplexity Computer in February 2026, a cloud orchestration system managing 19 AI models for complex tasks, available to premium subscribers. It has integrated with Samsung’s Galaxy S26, powering voice assistant functionality, and announced its first developer conference for March 2026. The injunction constrains one product in one jurisdiction; it does not alter the underlying commercial thesis.

That thesis, agentic AI as the primary interface for digital commerce, is not going away. The question is whether it develops within a framework of negotiated platform access or through continued legal conflict. For senior investors weighing positions in both incumbents and challengers, the answer will likely determine the trajectory of returns across the sector for the rest of this decade.