Detection Summary
This detection indicates that a JSON file distributed with a model contains a credential, such as a cloud access key, an API token, a webhook URL, or a private key, or a field whose name and value strongly suggest one. The detection evidence identifies the credential type and the field where it was found; the value itself is redacted in the report.Security Impact
A credential embedded in a model artifact is exposed to everyone who can download the artifact, including public repository users, registry mirrors, and any system that caches the file. Depending on the credential, an adversary may gain access to model repositories, training infrastructure, cloud storage, or third-party services, enabling data theft, model tampering, or lateral movement. The exposure persists in version history and downstream copies even after the file is fixed.False Positive Considerations
Structured credential matches (provider-prefixed tokens, private key blocks, webhook URLs) are reliable indicators. Suspected-secret matches on generic field names may occasionally flag non-secret values with credential-like shape; validate whether the flagged field holds a real credential before acting. Deliberately planted test or documentation values can also trigger this detection.Recommended Remediation
- Treat the credential as compromised: revoke or rotate it with the issuing service immediately, regardless of how long the artifact has been available.
- Remove the credential from the file and republish the artifact. Remember that repository version history and downstream mirrors retain the old file; deleting the current copy does not undo the exposure.
- Review access logs for the affected service or account for signs of unauthorized use since the artifact was first distributed.
- Identify how the credential entered the artifact (for example, training scripts serializing their configuration) and add pre-publication secret scanning to the model release process to prevent recurrence.