Amazon’s “transaction-entry” defense just got court-tested—AI shopping agents may need a new payments-and-permissions layer

[Amazon.com Services LLC v. Perplexity AI, Inc.] played out in federal court in California with Amazon seeking emergency relief against Perplexity’s Comet “agentic” shopping tool.

On March 9–10, 2026, U.S. District Judge Maxine M. Chesney granted a preliminary injunction that temporarily blocks Perplexity from using AI agents (deployed through the Comet browser) to access Amazon’s protected computer systems and to use/take over Amazon accounts for that purpose. The order also defines “AI agents” as software/program deployed through Comet that can autonomously or semi-autonomously interact with third-party sites on behalf of a user.

The court’s likely-success analysis centers on authorization: the order indicates Amazon can likely prove access “with the Amazon user’s permission but without authorization by Amazon,” including obtaining private account information transmitted to Perplexity’s servers.

In this dispute, Amazon’s injunction language targets interface/entry and identity/authorization boundaries: access to protected systems and account-scoped operations by AI agents using Comet. That’s why the “first round” matters for long-run market structure: it determines whether agentic tools have a sanctioned integration path (API/partner rails) or must operate in a discovery-only mode.

Amazon’s advantage is that it already spans multiple commerce layers—retail interface, merchant platform capabilities, and large-scale cloud infrastructure—so it can potentially offer compliant integration while excluding unsanctioned access.

The order (and related reporting of its key reasoning) describes evidence consistent with: (1) access to password-protected sections and (2) retrieval of private Amazon account information, which is then transmitted to Perplexity servers to perform requested tasks.

It also references statutory framing including a loss threshold element under the CFAA structure (the order includes language about loss aggregating at least $5,000 over a one-year period). The injunction therefore isn’t only about “automated behavior.” It’s about the combination of protected access + account data + unauthorized-by-platform authorization.

The “new analytical question” in your brief is exactly the right lens: whoever controls the transaction-entry layer can become the profit center, even if they lose some discovery credit. This court setback gives Amazon leverage to shape how agentic commerce enters its environment—potentially via permissioned integration rather than scraping-like access.

For markets, that implies two competing scenarios: 1) Discovery-only mode dominates in the near term, reducing conversion rates and tightening monetization for agentic tools. 2) Permissioned rails expand in the medium term, where winners are firms providing the sanctioned interface, identity, and checkout pathways.

Amazon.com, Inc. also reported large revenue scale and positive EBIT margin in its latest snapshot metrics from the data tool, indicating operating capacity to fund platform defenses while negotiating integration paths.

This matters because agentic access disputes are asymmetric: a discovery startup must redesign its agent and compliance posture; a platform owner can offer integration standards while seeking injunctions against conduct it deems unauthorized.

To operationalize this theme, investors can watch public companies that (a) provide the infrastructure behind sanctioned access, (b) monetize enterprise AI agent execution in ways that align with permissioning, or (c) provide commerce rails adjacent to platform checkout.