Protocols & news

Web Bot Auth: how AI agents sign their requests, and why it matters

Listen · 8 min

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

Short answer

Web Bot Auth is a way for an AI agent to prove who it is. The agent signs each request with a private key, and your site (or your CDN) checks the signature against a public key the agent's operator publishes. It is stronger than trusting a user agent or an IP address, and it is still an IETF draft, not a final standard.

Key takeaways

  • Signed requests prove identity far better than user agents or IP addresses.
  • Signatures are short-lived, so they must be checked live, request by request.
  • Web Bot Auth is still an IETF draft; Cloudflare already verifies it.

Why are user agents and IP addresses weak proof of identity?

Because anyone can fake a user agent, and an IP address says little about who is using it.

A user agent is a line of text a visitor sends to say what it is. The visitor writes that line itself. Nothing stops a script from claiming to be Chrome, or to be a well-known AI agent.

IP addresses are better, but still fragile. Cloudflare's own write-up points out that one address can be shared by many users or services, such as VPNs, privacy proxies and shared cloud infrastructure. Cloud providers also change their address ranges over time, so an allowlist of IPs goes stale.

For a store owner, the result is simple. A line in your logs that says "ChatGPT" is a claim, not proof. Our guide on how to tell if a visitor is really ChatGPT covers the older checks. Web Bot Auth is the newer, stronger one.

What is Web Bot Auth?

Web Bot Auth is a method that lets an automated visitor prove its identity with a digital signature on each request.

It is not a new product. It is a way of using an existing standard, RFC 9421 (HTTP Message Signatures), for bots and AI agents. The IETF has a working group for it, called Web Bot Auth. Its charter covers crawlers, link checkers, AI training crawlers and AI agents acting for end users.

Think of it as a wax seal on a letter. Anyone can write "from the bank" on an envelope. Only the bank has the seal.

How does a signed request work?

The agent signs the request with a private key, and your site checks the signature with the matching public key.

In plain steps:

  1. The agent's operator creates a pair of keys. The private key stays secret. The public key is shared.
  2. The operator publishes its public keys in a key directory on its own website.
  3. For each request, the agent adds a signature made with the private key.
  4. The request also says where the public keys can be found.
  5. Your website, or a service in front of it, fetches the public key and checks the signature. If it matches, the request really came from that operator.

An attacker can copy the public key. That does not help, because only the private key can make a valid signature.

In technical terms, three headers carry this. Signature holds the signature. Signature-Input describes what was signed and when, including a creation time and an expiry time. Signature-Agent points to the key directory. Cloudflare's documentation shows the same three headers, and the current IETF draft describes Signature-Agent as a list of web addresses that each point to a published set of keys.

Signature-Agent: "https://agent.example"
Signature-Input: sig1=("@authority" "signature-agent");created=1700000000;expires=1700000060;keyid="...";tag="web-bot-auth"
Signature: sig1=:...:

Cloudflare's docs say bots serve their key set at /.well-known/http-message-signatures-directory. The IETF draft describes a well-known address for the directory too, though it does not give that exact path in the part we reviewed. The values above are illustrative, not from a real bot.

Why do signatures expire within minutes?

Because a short life limits the damage if someone copies a signed request.

Each signature carries a creation time and an expiry time, as RFC 9421 allows. Cloudflare's guidance is to keep the window short, often about a minute. Cloudflare also says it does not track which nonces it has seen, so the short expiry is what limits replay attacks. A nonce is a random one-time value a signer can add.

This has a practical effect for you. A signature cannot be checked once and trusted forever. Every request carries a fresh signature, and every request has to be checked as it arrives. A log entry from last week cannot prove anything about a signature made today. Checking has to happen live, at the edge or on your server.

Who signs requests today?

Some large operators already do, and Cloudflare verifies them. The list is changing, so check current documentation.

What we could confirm from primary sources:

  • OpenAI. OpenAI's developer guidance says its cloud browser (ChatGPT agent) signs requests with a Signature-Agent header set to https://chatgpt.com, and publishes its keys at https://chatgpt.com/.well-known/http-message-signatures-directory. OpenAI says this applies to signed cloud browser requests, not to its web crawlers. See our page on ChatGPT agent.
  • Bots in Cloudflare's verified bots program. Cloudflare says a bot can qualify by declaring its identity with a Web Bot Auth signature, or through a published IP list with a stable user agent, or reverse DNS. It also says the bot must behave: obey robots.txt, keep request rates reasonable and not evade owner preferences.

A verified bot is one whose identity has been confirmed and whose behavior has been reviewed. A signature handles the first part. It does not handle the second.

We have not confirmed a complete list of other signers, so we do not name any.

Is Web Bot Auth a final standard?

No. It is an Internet-Draft, which means work in progress.

The pieces differ in maturity:

  • RFC 9421 (HTTP Message Signatures) is a published Proposed Standard from February 2024.
  • draft-ietf-webbotauth-httpsig-protocol-01 is the current protocol draft we found, dated 29 October 2026. It is a working group document, intended for the standards track, and it replaces an earlier individual draft. Its listed authors are from Cloudflare and Google.
  • The IETF Web Bot Auth working group is active.

Drafts can change, and details such as header formats may shift. Cloudflare's docs also describe an experimental way to pass on the end user behind an agent, and say it may change. Build on what is stable, and expect updates.

What can a website owner do now?

Turn on a check that verifies signatures, then decide what verified agents may do.

  1. Use a service that verifies signatures. On Cloudflare, verification is handled for you, and you do not sign anything. Cloudflare sets a verified-bot field you can use in rules.
  2. Keep the signature headers intact. If you run proxies between the visitor and your site, make sure they pass through Signature, Signature-Input and Signature-Agent.
  3. Write rules by identity, not by guesswork. Allow, limit or block verified agents by name. Treat unsigned claims as unverified.
  4. Do not stop at identity. Decide what each agent may read and do, especially if it can place orders.
  5. Run a free check. Our readiness scan shows how your site looks to AI agents today.

How Sill helps

Sill shows you which AI agents and bots visit your site, and it verifies signed agent requests (Web Bot Auth) on Cloudflare and WordPress installs. You see which visitors proved who they are and which only claimed to.

Questions people ask

Is Web Bot Auth an official internet standard?

Not yet. It is an Internet-Draft in the IETF's Web Bot Auth working group, intended to become a standards-track document. It builds on RFC 9421, which is a published standard.

Do I need to sign anything as a website owner?

No. Bots sign their own requests. You only need something on your side, such as Cloudflare or a plugin, that checks the signatures.

Does a valid signature mean a bot is well behaved?

No. A signature proves who sent the request. It says nothing about whether you want that visitor. You still decide what each verified agent may do.

What happens to agents that do not sign requests?

They can still visit, but you cannot prove who they are. Treat their identity claims as unverified and apply your usual rules.

Sources

  1. OpenAI: Help ensure cloud browser access to your site · checked 15 Nov 2026
  2. IETF: draft-ietf-webbotauth-httpsig-protocol-01, HTTP Message Signatures for automated traffic · checked 15 Nov 2026
  3. IETF: Web Bot Auth (webbotauth) working group · checked 15 Nov 2026
  4. RFC 9421: HTTP Message Signatures · checked 15 Nov 2026
  5. Cloudflare docs: Web Bot Auth · checked 15 Nov 2026
  6. Cloudflare docs: Verified bots · checked 15 Nov 2026
  7. Cloudflare blog: Forget IPs: using cryptography to verify bot and agent traffic · checked 15 Nov 2026

Written by The Sill Signals Team. Last reviewed 15 Nov 2026.

Share this articleLinkedInXBlueskyEmail

Keep reading

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

Scan your site free