Trust & security

Can agents prove they're trusted without revealing who they are? Anonymous bot authentication

Listen · 10 min

Narrated by an AI voice. Code examples are on the page.

Short answer

Yes, in principle. Privacy Pass, published by the IETF in 2024, lets a trusted party vouch for a visitor with a token your site can check without learning who the visitor is. Proposals to apply the same idea to AI agents and bots are still early drafts, so today the practical way for an agent to prove it is trusted is to sign its requests with Web Bot Auth, which does reveal who it is.

Key takeaways

  • Privacy Pass lets a site check a vouch, not the visitor's identity.
  • Anonymous bot authentication for AI agents is still an early IETF proposal.
  • Web Bot Auth, which names the agent, is what agents use today.

Can an AI agent prove it is trusted without saying who it is?

Yes, in principle. A cryptographic tool called Privacy Pass lets a trusted third party vouch for a visitor, and lets your website check that vouch without learning who the visitor is. Applying it to AI agents and bots is still at the proposal stage, so no agent you meet today proves its trust this way.

The reason anyone wants this is a real tension. Store owners want to know which bots are well behaved. Agents act for people, and those people may not want every site to know which tool they use, or to have their visits linked together. Anonymous bot authentication tries to give the site the first thing without giving away the second.

How do agents prove who they are today?

Today, agents that want to be recognised sign their requests with Web Bot Auth. The agent's operator publishes a public key, and your site or CDN checks each request's signature against it. Our guide to how signed agent requests work covers the details.

A signature is the opposite of anonymous. It tells you exactly which operator sent the request, and that is the point. It lets you allow ChatGPT agent, say, while treating an unsigned request that only claims to be ChatGPT as unverified.

Cloudflare has written about the limit of that approach. In an October 2025 post it says Web Bot Auth "would allow agents to identify to the origin which platform they're operated by," but that it would not want to extend that mechanism to "identifying individual users of the platforms, as this would create an unacceptable privacy risk for these users." So a signature can tell you the platform. It should not tell you the person behind it.

What is Privacy Pass?

Privacy Pass is defined in three IETF RFCs published in June 2024: RFC 9576 (an informational architecture document) and the standards-track RFC 9577 and RFC 9578. RFC 9576 describes it as authorization "based on privacy-preserving authentication mechanisms." In plain terms: one party checks the visitor, another issues a token, and your site accepts the token without learning anything that identifies the visitor.

The architecture has four roles:

Role What RFC 9576 says it does
Client "An entity that seeks authorization to an Origin."
Attester "An entity that attests to properties of Clients for the purposes of token issuance."
Issuer "An entity that issues tokens to Clients for properties attested to by the Attester."
Origin "An entity that consumes tokens presented by Clients and uses them to make authorization decisions."

The Origin is your website. The privacy goal is that the Origin cannot tell whether two token redemptions came from the same Client. RFC 9577 defines how a site asks for a token over HTTP, using a PrivateToken challenge, and RFC 9578 defines two ways tokens are issued: one checked with the issuer's private key, one that anyone with the issuer's public key can check.

You may already have seen it work. In 2022 Cloudflare and Apple used the approach in Private Access Tokens to skip CAPTCHAs on iPhones and Macs. In that setup Apple is the attester, and Cloudflare says the website "knows only your URL and IP, which it has to know to make a connection."

How would anonymous bot authentication work for AI agents?

Anonymous bot authentication would let a bot show your site that it belongs to a group of approved bots, without showing which one it is. The main written proposal is an IETF draft titled "Anonymous Bot Authentication: Authorization and Rate Limiting for Web Agents," by Eric Rescorla and Richard Barnes. Version 01 is dated 19 July 2026.

The idea works like this:

  1. A bot registers with a credential "Anchor," which checks that it follows the Anchor's policies. The draft gives examples such as being a registered corporation or holding a domain name.
  2. The Anchor endorses the bot, and a separate party the draft calls a Moderator turns that endorsement into an anonymous credential.
  3. The bot presents that credential when it visits your site.
  4. Your site learns the bot is approved by that Anchor, but the credential does not tell it which bot. The draft notes a bot may still be identifiable by other means, such as its IP address or fingerprinting.

The draft puts it this way: "the site only learns that someone with a credential from the Anchor is authenticating, not whether it's Alice or Bob." Sites choose which Anchors to accept. The draft also lets each credential carry a usage limit, so a site can still cap how many requests any one bot makes.

Cloudflare has explored a related idea for agents. Its October 2025 post describes anonymous credentials that let site operators enforce policies "without ever having to identify any user or track them across requests," and says the work is "still in its early stages."

For a store owner, there is a trade. You would gain a way to let approved bots through without learning who is behind them, and without each bot needing its own entry on your allowlist. You would lose the ability to treat each bot individually.

The draft is direct about the trade-off. Using anonymous authentication, a site cannot:

  • selectively block one bot,
  • see which bots are reading which pages, or
  • link several visits by the same bot.

The draft says it does support rate limits per bot, separating bots from humans, and requiring bots to meet certain criteria. It says it does not support allow lists, deny lists or auditing bot behaviour. In its words, "abuse controls that rely on knowing the identity of the abusive party are incompatible with anonymous authentication."

For a store, that means anonymous credentials suit a different job from signatures. A signature answers "is this really ChatGPT agent?" An anonymous credential answers "is this one of the bots a trusted Anchor approved?"

Is any of this a standard yet?

Privacy Pass itself is a published standard. Anonymous authentication for bots is not. The Rescorla and Barnes draft is an individual submission with no working group adoption listed on the IETF Datatracker, and it opens with a disclaimer: "This is a work-in-progress draft and has not yet seen significant security analysis."

It is also separate from the Web Bot Auth working group's main work. That group's charter lists "Tracking or assigning reputation to particular bots" and "Authenticating the end user of a participating client or agent" as out of scope. Our article on where Web Bot Auth stands at the IETF covers that track.

So at the time of writing, the practical path is the one that names the agent. Cloudflare counts a Web Bot Auth signature as one way a bot can honestly identify itself on the way to becoming a verified bot, as our explainer on Cloudflare verified bots and signed agents describes.

What should you do now?

You do not need to change anything for anonymous bot authentication yet. Nothing in it is ready to switch on for a store. Three steps help today:

  1. Recognise agents that sign. Check signatures where your CDN or platform can, and treat unsigned claims as unverified. Our AI agent identity guide explains the methods.
  2. Rate-limit by behaviour, not by name. Limits on request volume protect you whether or not a bot identifies itself.
  3. Watch the drafts. If anonymous credentials mature, the choice of which Anchors to trust will matter as much as which bots you allow today.

This is general information about technical standards, not legal advice.

How Sill helps

Sill shows you which AI agents and bots visit your site and the pages they read. On Cloudflare and WordPress installs it verifies Web Bot Auth signatures from agents that sign, so a signed visit is marked as verified. Sill does not issue or check Privacy Pass tokens. You can see how your site looks to agents with a free readiness scan.

Questions people ask

Is Privacy Pass the same as a CAPTCHA?

No. Privacy Pass is a way to prove you already passed a check somewhere else, so a site can skip asking again. Cloudflare and Apple used it to cut CAPTCHAs on iPhones and Macs through Private Access Tokens.

Can I turn on anonymous bot authentication for my store today?

Not as a standard feature. The main proposal for bots is an individual IETF draft that its authors say has not yet had significant security analysis. Watch for it, but plan around signed requests for now.

Does anonymous authentication stop me blocking a bad bot?

It limits you. The draft says a site cannot use it to block one bot, see which bot read which pages, or link visits by the same bot. You can still rate-limit each bot and choose which vouching parties to accept.

Does Sill support Privacy Pass?

No. Sill verifies Web Bot Auth signatures on Cloudflare and WordPress installs, for agents that sign. It does not issue or check Privacy Pass tokens.

Sources

  1. RFC 9576: The Privacy Pass Architecture · checked 17 Nov 2026
  2. RFC 9577: The Privacy Pass HTTP Authentication Scheme · checked 17 Nov 2026
  3. RFC 9578: Privacy Pass Issuance Protocols · checked 17 Nov 2026
  4. Cloudflare: Private Access Tokens: eliminating CAPTCHAs on iPhones and Macs with open standards · checked 17 Nov 2026
  5. Cloudflare: Anonymous credentials: rate-limiting bots and agents without compromising privacy · checked 17 Nov 2026
  6. IETF: Anonymous Bot Authentication: Authorization and Rate Limiting for Web Agents (draft-rescorla-anonymous-webbotauth-01) · checked 17 Nov 2026
  7. IETF Datatracker: draft-rescorla-anonymous-webbotauth status · checked 17 Nov 2026
  8. Cloudflare docs: Verified bots · checked 17 Nov 2026
  9. IETF: Web Bot Auth (webbotauth) working group charter · checked 17 Nov 2026

Written by The Sill Signals Team. Last reviewed 17 Nov 2026.

Share this articleLinkedInXBlueskyEmail

Keep reading

See which AI agents visit your site, and decide what they may do.

Scan your site free