AI agents can act for people online, but websites may struggle to tell which person authorised an action or who is accountable when it goes wrong. Cloudflare and Tools for Humanity are proposing a framework to close that gap, with graduated identity checks rather than a blanket choice between blocking agents and asking everyone to reveal who they are.
Cloudflare Watch analysis
What happened
Cloudflare published its Trusted Agent Delegation initiative on 5 November. The proposal combines levels of user identification with reputation and accountability mechanisms. Cloudflare product director Bryan Becker and Tools for Humanity chief product officer Tiago Sada presented it jointly and invited feedback; they said a working group of agent builders, identity and payment providers was being formed, according to Biometric Update.
The framework describes four identification levels: no user identifier, a session-level identifier, an account-level identifier, and a persistent third-party identifier for higher-risk actions. The last could involve a government ID or a proof-of-human credential. The proposal also envisages websites building trust over time and agent operators acting on reports of abuse, instead of a site having to block an entire agent because of one user.
Why it matters
An agent can make a site’s usual signals, such as a device or browser tied to one person, much less useful. If several people’s requests arrive through the same agent, a business may have trouble distinguishing legitimate activity from abuse or limiting action against one account without affecting others.
The proposed tiers give websites a way to ask for more identification as the stakes rise. That could help with accountability, but persistent identifiers also raise questions about privacy and whether activity can be linked across services. This is a proposal, not a settled internet standard: the report says the coalition plans to take defined problems to the Internet Engineering Task Force, and gives no timetable for implementation.
Our read
The useful idea is the graduated approach: a low-risk request need not demand the same proof as a sensitive transaction. But the promise depends on how the framework balances accountability with privacy, and on whether agent operators actually respond when users abuse their systems. A tidy diagram is not a substitute for rules that work across services.
What to watch
- Whether the working group publishes technical details or a draft standard.
- How the proposal handles persistent identifiers and cross-service linkability.
- Whether agent operators adopt a clear process for investigating and limiting abusive users.
Discussion spark: For high-risk actions by AI agents, should websites be able to require a persistent identity credential, or should session-level identification be the privacy limit?
Sources and evidence
- Source update (8 November 2026, 17:30 UTC)
Independent WittyWires tracker for public updates about Cloudflare. Not affiliated with or endorsed by Cloudflare; this is not an official account.