Just posted to my Forbes column about a court ruling that could shape the future of AI agents and how they interact with major platforms.
A federal judge in the Northern District of California ruled that AI agents cannot access a company’s systems without the platform’s permission, even if the user themselves authorized the agent. The case centers on Amazon’s attempt to block Perplexity’s Comet agent, which can log into Amazon and shop on a user’s behalf. At the heart of the dispute is control. Amazon argues that agents “covertly intrude” into its systems and bypass its curated shopping experience, while Perplexity says its agents simply act with the same permissions a human user already has.
The judge suggested Amazon is likely to succeed in its lawsuit, writing that the company presented strong evidence that Comet accessed password-protected accounts without Amazon’s authorization and transmitted account information back to Perplexity’s servers to complete tasks.
If this reasoning ultimately holds, it could significantly reshape the agent economy. User permission alone would not be enough. Platforms would control whether AI agents can operate on their systems at all, meaning agents may need formal APIs, partnerships, or platform approval to function.