Cloudflare says websites need new rules for AI agents

In a post, Cloudflare says personal AI agents receive instructions from users and control a cloud computer or browser, usually through an operator. Examples it cites include Muse from Meta and dots from OpenAI. Because these agents can operate continuously, in parallel and at scale, they can remove limits that normally apply to human browsing; the company says agentic traffic may eclipse traditional human browsing on the internet in the near future.

Cloudflare compares these systems with cloud browsers and proxies: a website sees the intermediary rather than the user directly, losing information needed to recognize, serve and hold the person involved accountable. The company proposes an Intermediation Principle, with tools to preserve the relationship between websites and users and make agents and operators identifiable and accountable. The post focuses on agents hosted away from the user’s device and leaves local agents outside its primary scope.