GitGuardian: AI-assisted commits leak secrets at twice the rate

GitGuardian's 2026 State of Secrets Sprawl report found that commits flagged as AI-assisted leak credentials roughly twice as often as other commits. The cause is a workflow problem: generated code looks correct, and hardcoded keys in configs or test fixtures pass review and unit tests unnoticed.

• AI-assisted commits leak secrets about twice as often as others
• Secret scanning catches a credential only after it enters repo history
• Rotation and short-lived scoped credentials are the real remediation
• Coding agents should be instructed never to write credential values into source