Cloudflare and World have introduced Trusted Agent Delegation, a framework intended to let personal AI agents use websites under conditions those sites can assess. The proposal could help move agent access beyond blanket blocks, but it begins with hosted agents, not ones running locally on a user’s device.
Cloudflare Watch analysis
What happened
The framework sets out how a person could delegate tasks to an agent while a website sets the terms for that agent’s access. Its stated aim is to give sites a way to distinguish agents and assess whether to serve them, rather than treating every request from an agent operator’s infrastructure as interchangeable. World describes the proposal in its announcement and overview.
The overview points to several pieces that could support the approach: Web Bot Auth, which lets an agent sign its requests; protocols for agents and websites to negotiate; ways to provide claims such as whether a person is over 18 or is the cardholder; and machine-readable policies for what agents may do on a site. It also cites MoLE as an upcoming approach to rate limiting without user-specific identifiers, and work in the IETF’s aipref group on site policies.
The initial focus is agents hosted by an operator whose reputation a website can assess. World says on-device and locally hosted agents raise different issues around reputation and accountability, and are outside the framework’s first scope.
Why it matters
Personal agents may shop, compare prices or book travel on someone’s behalf. Websites need a way to decide whether those requests are welcome without either opening the door indiscriminately or shutting out an entire agent service. The proposed framework puts identity, permissions and accountability at the centre of that decision.
That is a meaningful problem to tackle, but this is a framework proposal, not evidence that websites and agent builders have adopted a shared system. The hard work will be agreeing what information is needed, who can provide it and how to avoid turning trust checks into a new obstacle for legitimate users.
Our read
The useful idea here is not simply “let agents in”. It is to let websites define conditions while preserving the link to the person who delegated the task. That could make agent access more workable for both sides, if the rules are interoperable and do not demand more personal data than the task requires.
For now, treat Trusted Agent Delegation as a proposed direction, not a ready-made access standard. The hosted-agent starting point is practical, but leaves the question of how websites should handle agents running on a person’s own device for later.
What to watch
- Whether Cloudflare, World and other participants publish a working implementation or interoperability tests.
- What identity and personal claims websites can request, and what safeguards limit unnecessary disclosure.
- Whether the proposal expands to locally hosted agents and explains how they can establish reputation and accountability.
Discussion spark: Should websites be able to require identity and accountability checks before serving a personal AI agent, or should agents be able to browse without tying their actions back to a user?
Sources and evidence
- Source update (12 November 2026, 00:00 UTC)
Independent WittyWires tracker for public updates about Cloudflare. Not affiliated with or endorsed by Cloudflare; this is not an official account.