Allow ChatGPT cloud browser in Cloudflare, Akamai, HUMAN, Vercel
How to allow ChatGPT cloud browser through your WAF: the exact Cloudflare, Akamai, HUMAN and Vercel settings, manual signature checks, and how to test the rule.
By Crawlito · 9 min read
In short
- OpenAI says ChatGPT's cloud browser does not need allowlisting by default, but security settings can block it, and Akamai, Cloudflare, HUMAN and Vercel verify its request signatures for you.
- On Cloudflare, OpenAI points to two options: set dots, OpenAI to Allow under Personal Agents in AI Crawler and Agent access policies, or write a WAF custom rule on detection ID 129220581.
- Cloudflare documents detection IDs as an Enterprise Bot Management feature, while its Free and Pro plan pages list the AI bot policies, so non-Enterprise sites should use the Personal Agents setting.
- On any other CDN or firewall, an exception should apply only after the Signature-Agent header equals "https://chatgpt.com" and the RFC 9421 signature verifies against OpenAI's published key directory.
- An exception for the cloud browser does not change how OAI-SearchBot, GPTBot or ChatGPT-User are treated; those crawlers need their own rules.
On this page
- What ChatGPT's cloud browser is, and what it is called
- How to allow ChatGPT cloud browser in Cloudflare
- How to allow it in Akamai
- How to allow it in HUMAN
- How to allow it in Vercel
- How to verify signatures yourself on other firewalls
- How to test that the exception fires
- What the exception does not cover
- Frequently asked questions
- Check the pages the agent will land on
To allow ChatGPT's cloud browser through your WAF, allow the verified identity your security vendor already assigns to it, never a header string. OpenAI's Agent Access guide says Akamai, Cloudflare, HUMAN and Vercel all verify the cloud browser's request signatures automatically, so on those platforms the fix is a dashboard setting or a single rule. On anything else you verify the signature yourself, and only then apply the exception.
OpenAI also says the cloud browser can already use your site and you do not need to add it to an allowlist. You only need an exception if your bot rules are blocking or challenging it, which typically shows up as a cart, booking or checkout flow that stops partway when a user asks ChatGPT to complete a task. How the signatures work is covered in Web Bot Auth: how signed requests verify AI crawlers, and how these visits look in your logs is in ChatGPT agent and Atlas in server logs. This post is the per-vendor recipe.
What ChatGPT's cloud browser is, and what it is called
OpenAI describes it as "a browser running on a computer in the cloud" that ChatGPT uses for some tasks to read pages, click buttons and interact with a site for a user. It identifies itself with Web Bot Auth, built on RFC 9421 HTTP Message Signatures. OpenAI says this guidance covers signed cloud browser requests, not web crawlers.
The awkward part is naming. The same traffic carries a different label in each vendor's product. OpenAI calls these the vendors' existing entries, and their names (Operator, ChatGPT Agent) match earlier OpenAI product names:
| Vendor | Name shown for the cloud browser | Identifier |
|---|---|---|
| Cloudflare (access policies) | dots, OpenAI, under Personal Agents | none shown |
| Cloudflare (WAF) | ChatGPT Operator, signed agent | bot tag chatgpt-agent, detection ID 129220581 |
| Akamai | ChatGPT Agent, in the AI Fetchers and Agents category | none shown |
| HUMAN | ChatGPT Agent, in Known Bots & Crawlers | none shown |
| Vercel | ChatGPT Agent entry | chatgpt-operator |
All names and IDs above come from OpenAI's guide at the time of writing. OpenAI's own FAQ on that page calls the feature "Cloud Browser/Dots", which matches the Cloudflare label. If a vendor renames the entry, search its directory for OpenAI rather than for one exact name.
How to allow ChatGPT cloud browser in Cloudflare
OpenAI gives two options. Pick one based on your plan and how your rules are built.
Option 1: AI Crawler and Agent access policies
- In the Cloudflare dashboard, select your domain and go to Security > Settings.
- Filter by Bot traffic and select AI Crawler and Agent access policies.
- Under Personal Agents, set dots, OpenAI to Allow (do not block).
- Select Save.
Cloudflare's AI bot policy documentation adds the context. Its policies are split into Search, Agent and Training behaviors. Personal Agents inherit the Agents policy by default, and changing the Agents policy updates every Personal Agent, so if you later edit the Agents policy, check that the OpenAI entry is still on Allow. Cloudflare recommends Allow (do not block) for Personal Agents. Its bots changelog records that this screen was renamed from Configure AI bot policies on November 4, 2026, so older guides may use the old name. The wider settings page is covered in the Cloudflare AI crawler settings guide.
Option 2: a WAF custom rule on the detection ID
OpenAI's steps: open Security > WAF, create a custom rule, set the expression to Bot Detection ID equals, search for ChatGPT Operator and select 129220581, then set the rule to allow or skip matching requests and deploy it. Cloudflare's current docs place custom rules on the Security rules page, so your dashboard may show that name instead. In the expression editor, the field Cloudflare documents for this is cf.bot_management.detection_ids, which is an array:
any(cf.bot_management.detection_ids[*] eq 129220581)Four caveats from Cloudflare's own documentation:
- Plan. Cloudflare's detection IDs page says detection IDs are available to Enterprise Bot Management customers. On other plans, use Option 1.
- Action. Cloudflare's list of rule actions has no Allow action for custom rules (Allow belonged to the deprecated Firewall Rules), so in practice "allow" here means a rule with the Skip action.
- What Skip can skip. A Skip rule can skip the remaining custom rules, rate limiting rules, managed rules, Super Bot Fight Mode rules and some older products such as Browser Integrity Check. It cannot skip Bot Fight Mode, the Free plan feature.
- Order. The custom rules option skips the remaining rules in the ruleset, so place the Skip rule above the custom rules that would otherwise challenge the request.
How to allow it in Akamai
Akamai lists the traffic as ChatGPT Agent in its AI Fetchers and Agents category and verifies signatures itself. OpenAI's steps are:
- In your Akamai Security Configuration, open Shared Resources, then Custom Bot Categories.
- Select the + icon and add a new bot category, then save it.
- In that category, create a bot with Type Verified Bots and Bot name ChatGPT Agent.
- Close Shared Resources, open the relevant Security Policy, go to Bot & Agent Control > Custom Bots, and set the new category's action to Allow.
The custom category is the point: it lets you allow this one agent without changing the action for everything else in AI Fetchers and Agents. OpenAI adds that Akamai verifies the signatures, so no custom verification is needed.
How to allow it in HUMAN
HUMAN verifies the signatures too, and OpenAI documents two places to look.
- HUMAN Sightline. In your Sightline or BD console, go to Policies > Traffic Policy Settings > Known bots & Crawlers, search for ChatGPT Agent, set the rule to ON and select Allow.
- HUMAN AgenticTrust. In the Sightline console, open Policies > AI Agents Permissions, search for ChatGPT Agent and review the permissions. OpenAI notes these permission settings do not change what the cloud browser is able to do.
How to allow it in Vercel
On Vercel there is usually nothing to configure. OpenAI says Vercel recognizes the traffic through its ChatGPT Agent entry, identified as chatgpt-operator, and verifies the signatures automatically. Vercel's verified bots directory lists chatgpt-operator as an ai_assistant that handles user-initiated requests, and Vercel's August 2025 changelog says signed verified bots are allowed through Bot Protection and Challenge Mode.
The blocks on Vercel come from rules you wrote:
- WAF custom rules. Vercel says you can block verified bots with custom rules that match on the User Agent or the
Signature-Agentheader. Review any deny or challenge rule that could match this traffic. - A proxy in front. Vercel says Bot Protection does not work when a reverse proxy such as Cloudflare is placed in front of the deployment. In that setup, the proxy is where you allow the agent.
If the Vercel Security Checkpoint appears for other bots, check its rules for this one as well.
How to verify signatures yourself on other firewalls
If your CDN or firewall does not recognize the cloud browser, OpenAI's guide gives the manual path:
- Confirm the
Signature-Agentheader exactly matches"https://chatgpt.com", quotation marks included. - Fetch the public key from
https://chatgpt.com/.well-known/http-message-signatures-directory. - Verify the
SignatureandSignature-Inputheaders according to RFC 9421. - Apply the exception only to requests whose signatures verified.
OpenAI does not publish which message components its signatures cover. The request below is illustrative, with placeholder values, and follows the Web Bot Auth format in Cloudflare's documentation:
GET /checkout HTTP/1.1
Host: shop.example
Signature-Agent: "https://chatgpt.com"
Signature-Input: sig1=("@authority" "signature-agent");created=1700000000;expires=1700000060;keyid="KEY-THUMBPRINT";tag="web-bot-auth"
Signature: sig1=:BASE64-SIGNATURE:In practice, steps 2 and 3 run in edge code, such as a Worker or a reverse proxy plugin, rather than in the firewall rule itself. Cloudflare maintains an open source web-bot-auth repository with TypeScript and Rust packages for RFC 9421 and Web Bot Auth, and lists verification examples for Workers and a Caddy plugin; its README notes the code has not been audited. Whatever you use, have it check the created and expires window and that @authority is your own host, and reject expired signatures (the Web Bot Auth guide explains why short expiry is the replay defence). Then have the verifier set a value your firewall trusts, and allow on that value.
What never to do is allow on step 1 alone. The header is text, and anyone can send it.
How to test that the exception fires
You cannot sign a request as ChatGPT, so testing has two halves.
Prove spoofing fails. Send the header without a valid signature and confirm your rule does not let it through:
curl -s -o /dev/null -w "%{http_code}\n" \
-H 'Signature-Agent: "https://chatgpt.com"' \
-A "Mozilla/5.0" https://shop.example/checkoutYou should get the same response as any other unverified client. If this request bypasses your bot rules, the exception is keyed on the header, not on verification, and needs fixing.
Prove real traffic passes. Ask ChatGPT to complete a task on your site, then look at your vendor's logs:
- Cloudflare: in Security Events or Security Analytics, filter for the rule or the detection ID. Skip rules log matching requests by default; Cloudflare says that if you clear Log matching requests, those requests no longer appear in Security Events.
- Akamai, HUMAN, Vercel: check the bot or firewall logs for the ChatGPT Agent or
chatgpt-operatorentry and the action applied.
Test the whole task, not just the first page. A WAF challenge page on a later step, such as an API call from the checkout page, can still stop the task.
What the exception does not cover
OpenAI is explicit that an allow rule for verified cloud browser visits does not change how your site handles other traffic. The cloud browser is separate from OpenAI's crawlers:
| Agent | Purpose | How you allow it |
|---|---|---|
| ChatGPT cloud browser | Completes tasks for a user | Signed requests, rules above |
| OAI-SearchBot | Surfaces sites in ChatGPT search | User agent plus published IP ranges, robots.txt |
| ChatGPT-User | Fetches pages a user asks about | User agent plus published IP ranges |
| GPTBot | Crawls content that may be used for model training | User agent plus published IP ranges, robots.txt |
The user agents and IP lists are on OpenAI's crawler overview. Each crawler is covered in GPTBot, OAI-SearchBot and ChatGPT-User explained. Two more limits from OpenAI's FAQ: an exception can apply site-wide, so keep your page-level access controls, and site owners cannot yet control which actions users take, although OpenAI says it requires human review before any consequential action. Allowing the browser removes a technical barrier. It does not mean ChatGPT will recommend your products.
Frequently asked questions
Do I need to allowlist ChatGPT's cloud browser?
Not by default. OpenAI says it can already use your site, and you only need an exception if your CDN or firewall settings block verified visits.
Is it safe to allow requests with Signature-Agent set to chatgpt.com?
Only after the signature verifies. The header alone is a claim anyone can copy, so allow on your vendor's verified classification or on your own verifier's result.
Can I limit the cloud browser to certain pages?
OpenAI says an exception can apply site-wide and recommends keeping existing page-level access controls in place. Scope your allow rule by path if you want it narrower.
Does allowing it also allow OAI-SearchBot?
No. OpenAI says cloud browser visits are separate from its web crawlers, including OAI-SearchBot, which you control with robots.txt and its own user agent and IP rules.
Check the pages the agent will land on
An exception only helps if the pages behind it are readable. Run the free check on a product or booking page to see whether AI crawlers can fetch it and read its content, with no account needed. For a review of every edge rule and template across up to 500 pages, with evidence and prioritized fixes, see the audit.
Filed under CDN and bot protection. Topics: chatgpt, ai-agents, waf, web-bot-auth, bot-protection