Recover a Leaked Cloudflare Access Service Token
Use Cloudflare's recovery sequence and verify that every caller has picked up the rotated service token secret.

One CI job succeeds after a service token rotation. An hourly sync still fails on its next run. If both clients share a Cloudflare Access token, one successful request only verifies the client that made it.
In November 2026, GitHub added detection for Cloudflare Access service token secrets. Cloudflare says it disables an affected token when GitHub reports a leak in a public repository and sends a notification. The recovery procedure has three separate actions: rotate the secret, update the clients, and re-enable the token. GitHub detector announcement and Cloudflare partner reporting announcement
Capture the incident reference and list each caller
Open the Access administration view and confirm the exact token and status. For a reported leak, retain the alert time and source reference in the incident record. Cloudflare documents that rotation clears the token’s displayed leak metadata and points operators to admin activity logs for the audit history. Keep secret values out of the record. Service token recovery documentation
Make a row for every client that loads the credential, including jobs that run infrequently. List the secret-store reference, the person or team responsible for updating it, and whether the client reads it at startup, when a job is created, or on each request.
| Example caller | What to update | Recovery evidence |
|---|---|---|
| Release pipeline | Its CI secret and any job snapshot created before the change | A fresh job makes an authorized request to the intended protected application. |
| Scheduled sync | Its secret-store reference and any process holding the previous secret | A fresh execution authenticates and reaches the expected read-only route. |
| Long-running integration | The runtime secret and the process that loads it | The restarted or refreshed instance authenticates from its own runtime context. |
Rotate, update, then turn the token on
- Rotate the leaked token’s secret using the documented recovery action. The previous secret is revoked immediately and the Client ID stays the same.
- Update each client through its approved secret store. Refresh a running process if it keeps the previous value in memory.
- Re-enable the token in Access. Rotation leaves a token disabled if the partner leak report disabled it.
- Remove the exposed value from the reported location and handle any repository cleanup under your incident process. A removed value may already have been copied, so credential recovery includes rotation.
Confirm recovery from each client
Run a fresh authorized request through the same authentication configuration the real job uses. Where possible, use a read-only route that only returns its known response to an authorized client. Record the timestamp, expected status, and a non-sensitive response marker. A response from a cached or existing session may say little about whether the client loaded the replacement credential. Cloudflare describes when a service token must be sent with each request and how an Access session can be used. Service token request behavior
If one client still fails, compare the non-secret token identifier, the secret-store reference, whether the process reloaded it, and the Access policy for that application. A working sibling client makes that side-by-side check faster. Close the recovery record with an outcome for each caller, including inactive or retired clients whose credentials were removed.
References
- Cloudflare service token recovery gives the rotation sequence and leak metadata behavior.
- Cloudflare partner leak reporting describes automatic detection and disablement.
- GitHub secret scanning detector identifies the supported Cloudflare secret type.