GitHub previews push protection for AI-detected secrets beyond fixed patterns
GitHub announced a public preview extending push protection to AI-detected secrets on October 28. A classifier developed with Microsoft Applied Sciences looks at how potential credentials are used in surrounding code, rather than relying only on recognizable token formats.
Once enabled, the classifier runs alongside pattern-based detectors for new pushes, commits made in the GitHub interface and file uploads. GitHub says it rejects a detected change and reports the secret location. This does not mean it can identify every credential or guarantee that no secret will enter a repository.
The feature is off by default. A repository must already have both push protection and AI-detected secret scanning enabled before the new blocking feature can be enabled. Organization and enterprise security configurations can apply those settings to selected repositories.
The preview is available on github.com to organizations with Secret Protection or Advanced Security. Opting in and enabling the preview begins AI-credit billing. Post-push AI-detected secret alerts remain included with those licenses at no additional charge, which is a separate billing treatment from prevention at push time.
GitHub reports that early results more than doubled the number of pushes where a secret was caught, with user-reported precision comparable to issuer-defined detectors. Those are early vendor-reported results, not an independent accuracy guarantee or a finding that every blocked string is an active credential.
Existing bypass and delegated-bypass workflows remain in place. GitHub says agents must obtain authorization before bypassing or working around a block. Detecting a suspected secret, removing it from the proposed change and deciding whether an exposed credential needs rotation remain distinct review steps.