Invalid secrets no longer block your developers and AI coding agents
Release Date: November 20, 2026
![]()
ggshield scans now drop the secrets GitGuardian checked and found invalid. Until now, every scan reported each secret it found, whatever its validity, so an AI Hook interrupted the agent and a git hook refused the commit on a key that is invalid. When that happens a few times a day, developers turn the hooks off, leaving them with no prevention layer active.
When GitGuardian checks a secret against its provider and the provider says it no longer works, the scan API drops it from the result: the agent is not interrupted, the commit goes through, the CI job passes. Valid secrets still block, and so do the secrets GitGuardian cannot check, so a generic password or a credential whose provider is unreachable gets the same treatment as before. The filter runs on GitGuardian's side, so it applies to every ggshield version already installed, with nothing to upgrade.
What does this mean for you?
- Fewer interruptions, same protection: the only secrets that stop blocking developers and agents are the ones GitGuardian confirmed invalid with the provider. On our own developer fleet, that is about one block in eight.
- One behavior everywhere: AI Hooks,
pre-commit,pre-push,ggshield secret scan ciand your own calls to the/v1/scanand/v1/multiscanendpoints follow the same rule. - The dashboard still shows invalid incidents: your VCS integrations scan commits on their own, so a commit the hook let through because its secret is invalid still creates an incident on the GitGuardian Dashboard, automatically ignored as they are picked up by the "Invalid secrets" playbook.
Why is this important?
Each block on a secret that no longer works teaches developers that the hook raises false alarms, and the next real block will get bypassed. With AI coding agents, the cost is higher: an agent blocked on an invalid secret cannot rotate it or remove it from the environment, so it suggests disabling the hook or ignoring the secret. Keeping the block for secrets that are valid or unverified, and only those, keeps it credible.
Get started today!
Enabled by default on every SaaS workspace, as long as validity checks are on. To keep invalid secrets in your scan results, contact your Customer Success Manager. See Skip invalid secrets in ggshield and API scans.
Enhancements
- Perimeter: Paste a repository URL in the Perimeter search or in the Source filter to find the matching source, including a URL copied from a sub-page of the repository such as a file or a branch view.
- Team Sync: Members invited by a GitHub team sync are now attached to their teams while the invitation is pending, so they join them as soon as they accept instead of waiting for the next sync. The Notify members option now also controls the invitation email.
- Secret revocation: Until now, a secret revocation refused by the provider for lack of permission only reported "Revocation failed". The message now says the permission is missing and tells you to revoke the secret from the provider.
Fixes
- NHI Governance:
- Fixed an issue where the Publicly leaked policy was missing from the policy filter of the Identities page on workspaces without Public Monitoring, although the policy also fires on leaks found by HasMySecretLeaked.
- Fixed an issue where the graph of an incident with many occurrences could time out.
- Gerrit Integration: Fixed an issue where repositories with anonymous read access were classified as private.
- Integrations: Fixed an issue where a free workspace got the trial dialog instead of the install flow when installing a GitHub, GitLab, Bitbucket or Azure DevOps integration.















