Cloudflare says websites need ways to identify, assess and hold accountable AI agents acting on people’s behalf, without making every agent behave like a person. Its proposal matters because an agent can act at scale while a website may struggle to tell who it represents or who answers when something goes wrong.
Cloudflare Watch analysis
What happened
In its post on trusted agent delegation, Cloudflare argues that agents can obscure the relationship between a website and the person behind a request. A hosted agent might serve several users, leaving a site to see the intermediary rather than the person it is acting for. Cloudflare says this complicates decisions about trust, fraud, access and accountability.
The company argues that CAPTCHAs and human approval prompts do not solve the central problem: they can show that someone is present, but not establish whose request is being carried out or who is responsible for it. Broad access permissions have a similar shortcoming. They may show what an agent is allowed to do, Cloudflare says, but not who should answer for its actions.
Cloudflare points to a mix of existing and developing approaches. Web Bot Auth can let an agent sign requests to identify itself and, implicitly, its operator. The company also cites MOLE as a developing approach to rate limiting without user-specific identifiers, and OAuth as a way to convey account identity. Its proposed direction is graduated disclosure: some interactions may need no identifying information, while others may require a session, account or more persistent identity.
Why it matters
This is a practical design problem, not just a question of whether agents can complete online tasks. If websites cannot distinguish one user’s activity from another’s, they may have a harder time providing access, managing risk and deciding who is accountable. The mechanisms used to solve that problem will also shape how much personal information gets revealed.
Cloudflare’s argument is that sites and agents need protocols for communicating identity, reputation and policy. It is a proposal for how the web might adapt, not evidence that one settled standard or implementation has already solved the problem.
Our read
The useful insight is that “add a CAPTCHA” and “give the agent permission” are not complete answers to delegated access. The harder question is what information a site needs for a particular action, and how to provide it without turning every agent interaction into an identity check.
Cloudflare is making a case for standards and graduated disclosure. That is worth a serious discussion, especially because the company also has a commercial stake in how websites manage traffic and trust. Privacy-preserving options will need to be more than a reassuring label: sites, operators and users will need a clear account of what is disclosed and who can be held responsible.
What to watch
- Whether Web Bot Auth or other standards gain adoption among agent operators and websites.
- How proposed mechanisms such as MOLE balance rate limiting with user privacy.
- Whether sites can request different levels of identity for different actions, rather than defaulting to broad disclosure.
Discussion spark: When an AI agent acts for you online, what should a website be allowed to learn about you before it grants access?
Sources and evidence
- Agents aren’t humans (but they kinda are) (5 November 2026, 17:03 UTC)
Independent WittyWires tracker for public updates about Cloudflare. Not affiliated with or endorsed by Cloudflare; this is not an official account.