THE CRUNCH
Cloudflare has published an argument about what happens to trust on the web when AI agents do the browsing. Agents, automations that take instructions from a human and pursue goals by controlling a cloud computer or browser, are intermediaries: the site sees the agent, not the person behind it. That breaks the usual trio of identity, reputation and accountability that sites use to answer who is acting, whether they are trustworthy, and who answers for the action.
The post illustrates the problem with an analogy: a personal assistant named Ezra who runs errands for several clients. A grocer can no longer tell whose preferences are whose, a store cannot prove who authorised a disputed purchase, and a bank watching password guessing can only ban the assistant, which also bans her innocent clients. Swap Ezra for an agent and the businesses for websites, the post says, and you have the web today.
Cloudflare's proposed fix is what it calls the Intermediation Principle: the business relationship between users and sites, and each participant's security responsibilities, should hold up regardless of intermediation. Sites would need ways to recognise and hold accountable a user acting through an agent, while agents and the operators behind them would also need to be recognisable, because agents can scale human access almost without limit.
The stakes come from scale. Cloudflare argues agentic traffic is growing rapidly and may eclipse traditional human browsing in the near future, at which point simply blocking intermediary traffic stops working. It notes the argument covers remotely hosted agents such as Meta's Muse and OpenAI's dots, with local agents running on a user's own device out of scope.