Trust & security

Web Bot Auth at the IETF: where the standard stands in late 2026

Listen · 12 min

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

Short answer

Web Bot Auth is not a finished standard yet. The IETF working group adopted its protocol draft in September 2026, and the current version, draft-ietf-webbotauth-httpsig-protocol-01, is dated October 29, 2026. The charter aimed to send it for approval by April 2026, so it is running late, and early adopters such as Cloudflare and Google document different versions of the format.

Key takeaways

  • Web Bot Auth is a working group draft, not yet an RFC.
  • The charter's April and August 2026 milestones have passed.
  • Verifiers and signers document different header formats today.

Is Web Bot Auth an official standard yet?

No. At the time of writing, Web Bot Auth is an Internet-Draft owned by an IETF working group. That is a later stage than a personal proposal, but it is not yet a published standard.

The piece it builds on is finished. RFC 9421, HTTP Message Signatures, was published as a Standards Track RFC in February 2024. Web Bot Auth adds the parts bots need on top: the Signature-Agent header, a key directory format and a fixed web address for that directory. Those parts are still being written.

If you want to know how signing works, our explainer Web Bot Auth: how AI agents sign their requests covers it. This article is about where the standard itself stands, and what that means for your site.

What is the current Web Bot Auth draft?

The working group's one active document is draft-ietf-webbotauth-httpsig-protocol-01, titled "HTTP Message Signatures for automated traffic" and dated October 29, 2026.

Some facts from the IETF Datatracker and the draft itself:

  • Authors. Thibault Meunier (Cloudflare) and Sandor Major (Google).
  • Stage. It is a "WG Document." Its IESG state is "I-D Exists," which means it has not been sent to the IETF's steering group for approval.
  • Aim. The draft's header says it is intended for the Standards Track.
  • Expiry. It expires on May 2, 2027, unless a newer version replaces it first.

The draft's abstract says it "defines the Signature-Agent header field for in-band key discovery, a key directory format based on JWKS, and a well-known URI at which that directory is served."

How did the Web Bot Auth standard get here?

It began as individual drafts, gained an IETF working group in 2025, and became a working group document in September 2026.

The timeline we could confirm on the Datatracker:

  • October 23, 2025. The IETF approved the working group's charter. The charter says the group "will standardize methods for cryptographically authenticating automated clients and providing additional information about their operators to Web sites."
  • March 2, 2026. Two earlier drafts, one on the architecture and one on the key directory, reached version -05. Both were later replaced.
  • June 26, 2026. A new individual draft, draft-meunier-webbotauth-httpsig-protocol, appeared. Its next version, in August 2026, folded the key directory draft into it and moved it to the Standards Track.
  • September 1, 2026. The working group adopted it as draft-ietf-webbotauth-httpsig-protocol-00.
  • October 29, 2026. Version -01 was published.

The working group met at IETF 127 in San Francisco, at 00:30 UTC on November 18, 2026. The published minutes say the group reviewed the adopted protocol draft, implementer feedback and several proposals that are not adopted.

What the minutes record as decisions and action items:

  • Signature tagging. The authors will replace the uncoordinated tag=webbotauth parameter with a defined signature parameter or a required set of signed headers.
  • Minimal signed components. The authors will specify a baseline set of signed components: authority or host, signature-agent and freshness. The aim is to keep signatures cacheable and reusable.
  • Error reporting. The authors will draft standard response headers that report signature verification failures, at the SHOULD requirement level.
  • Centralization. The authors will add text on centralization risks and the position of independent crawlers to the security and privacy considerations.
  • Scope. The minutes say tracking or reputation for sub-agents, sessions or end users is "explicitly out of scope" under the current charter. They also say non-cryptographic IP range discovery is outside the charter, and the chairs will consult the Area Director on next steps.

The minutes say the authors will publish an updated revision of the draft with the changes on signature tagging, baseline signed components and error headers. At the time of writing, the Datatracker still shows version -01 as the latest. These are minutes of a meeting, not a published specification, so details may change.

Is the Web Bot Auth working group behind schedule?

Yes, against its own charter. The charter set three milestones, and all three dates have passed:

Milestone in the charter Due
Standards track specification(s) describing authentication technique(s) sent to the IESG April 2026
Standards track specification(s) describing a means for conveying additional information about bots sent to the IESG April 2026
Best Current Practice operational specification sent to the IESG August 2026

None of these has happened yet. The protocol draft was only adopted in September, and the Datatracker shows no working group document for the operational guidance. Slipping milestones is common at the IETF, so this is a sign that the work is still in progress, not that it has stalled.

The charter also limits what Web Bot Auth will do. It covers search crawlers, archivers, link checkers, AI training crawlers and "AI agents retrieving or interacting with content on behalf of end users." It leaves out HTTP APIs and agent-to-agent interfaces, and it says "authentication of the end user is out-of-scope." Web Bot Auth tells you which operator sent a request, not which person asked for it.

What changed in the latest Web Bot Auth draft?

Version -01 made several changes that matter to anyone who verifies signatures. From the draft's change log:

  • Every Signature-Agent value is now a direct link to a set of keys. Other directory types were removed.
  • A site that expects signatures should ask for them with the Accept-Signature header from RFC 9421. At IETF 127, according to the minutes, several participants argued that servers should not be able to arbitrarily force clients to sign dynamic per-request elements, such as paths or nonces, so this part may change.
  • An operator that wants sites to tell its crawler apart from its user-directed agent should use separate Signature-Agent URLs, keys and key directories for each.

The draft is also blunt about what a valid signature does not prove. It "says nothing about who operates the Agent, whether the Agent is benign, or whether the request is authorized. Those are origin policy." In other words, identity is the standard's job, and deciding what to allow stays with you.

Around the main draft, the Datatracker lists more than ten related individual drafts. Recent examples include a registry and "Signature Agent card" format (October 27, 2026), "A JSON-Based Format for Publishing IP Ranges of Automated HTTP Clients" (October 22, 2026), a "Signed-Agent-Metadata" header (November 2, 2026) and a draft on attested conduct for anonymous web data collection (November 16, 2026). These are proposals. None has been adopted by the working group. Our agent registry entry covers the registry idea.

Do Cloudflare, Google and OpenAI follow the same draft?

Not exactly. Each documents a different stage of the work, so the header format you see depends on who signed the request.

  • Cloudflare says its implementation follows an older directory draft, version -03, where Signature-Agent is a plain quoted string such as "https://signature-agent.test". Its docs say verification fails if a message "uses the dictionary form from later drafts."
  • Google calls its own Web Bot Auth support "experimental" and says signed Google-Agent requests carry Signature-Agent set to g="https://agent.bot.goog". That is the labelled dictionary form from the newer drafts.
  • OpenAI says ChatGPT agent's signed requests carry Signature-Agent set to "https://chatgpt.com", the older quoted form, and asks sites to match it "including the quotation marks."

The current draft allows for this. It tells signers to send the dictionary form and says verifiers "MAY accept" the older form, which "deployments still send." We could not confirm how Cloudflare handles Google's signed requests in practice, so we don't draw a conclusion. What it shows is that a standard still in draft can be read differently by different companies.

Google also recommends that "in addition to Web Bot Auth you continue relying on IP addresses, reverse DNS, and user-agent strings" while it rolls out signed traffic. The draft says much the same about older checks during rollout. See our AI agent identity guide for how those checks compare.

What should a website owner do while Web Bot Auth is a draft?

Use it, but don't rely on it alone.

  1. Verify signatures where you can. A valid signature is still the strongest identity check available for the agents that sign. On Cloudflare, see Cloudflare signed agents and verified bots.
  2. Keep your older checks. Many agents don't sign at all. Google and the draft both recommend keeping IP, reverse DNS and user-agent checks during the rollout.
  3. Treat a failed or unreadable signature as unverified, not as hostile. Formats are changing, and a mismatch may come from an older or newer draft.
  4. Expect changes. Header formats and directory rules have changed between drafts and may change again before an RFC.

Sill verifies Web Bot Auth signatures on Cloudflare and WordPress installs, for agents that sign, and shows you which visitors proved who they are and which only claimed to. Run a free readiness scan to see how your site looks to AI agents.

Questions people ask

When will Web Bot Auth become an RFC?

No date is set. The working group's milestones for sending its specifications to the IESG were April and August 2026, and those have passed. After that step, an IETF draft still goes through review and approval before it is published as an RFC.

Who wrote the Web Bot Auth draft?

The current working group draft lists two authors: Thibault Meunier of Cloudflare and Sandor Major of Google. As a working group document, it is now owned by the IETF working group, not by either company.

Can I join or follow the IETF Web Bot Auth work?

Yes. IETF work is open to anyone. The working group's mailing list is web-bot-auth@ietf.org, and its drafts, meetings and charter are public on the IETF Datatracker.

Does Web Bot Auth identify the person using an AI agent?

No. The working group's charter says authentication of the end user is out of scope. Web Bot Auth identifies the agent or bot operator, not the person behind a request.

Sources

  1. IETF Datatracker: Web Bot Auth (webbotauth) working group and charter · checked 17 Nov 2026
  2. IETF Datatracker: webbotauth working group documents · checked 17 Nov 2026
  3. IETF Datatracker: webbotauth meetings · checked 17 Nov 2026
  4. IETF Datatracker: Minutes IETF127: webbotauth · checked 19 Nov 2026
  5. IETF Datatracker: IETF 127 webbotauth session · checked 19 Nov 2026
  6. IETF Datatracker: Web Bot Auth (webbotauth) group history · checked 17 Nov 2026
  7. IETF: draft-ietf-webbotauth-httpsig-protocol-01, HTTP Message Signatures for automated traffic · checked 19 Nov 2026
  8. IETF: draft-ietf-webbotauth-httpsig-protocol-01 (full text) · checked 17 Nov 2026
  9. RFC 9421: HTTP Message Signatures · checked 17 Nov 2026
  10. Cloudflare docs: Web Bot Auth · checked 17 Nov 2026
  11. Google: Authenticate requests with Web Bot Auth (experimental) · checked 17 Nov 2026
  12. OpenAI: Help ensure cloud browser access to your site · checked 17 Nov 2026

Written by The Sill Signals Team. Last reviewed 19 Nov 2026.

Share this articleLinkedInXBlueskyEmail

Keep reading

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

Scan your site free