AI-Assisted Code Leaks Credentials at Double the Rate, Report Finds
GitGuardian's 2026 State of Secrets Sprawl Report shows AI-assisted commits leak secrets about twice as often as human-written ones. Most fast-growing categories of leaked credentials are now tied to AI. The report highlights how AI coding agents accelerate credential exposure.
A new industry report from GitGuardian examining credential exposure patterns finds that code commits generated with AI assistance are roughly twice as likely to contain leaked secrets compared to commits written by humans. The findings point to a notable shift in how sensitive information enters codebases.
The report's data indicates that the fastest-growing categories of exposed credentials are now predominantly associated with AI tooling. This trend suggests that AI coding agents, while boosting developer productivity, are simultaneously accelerating the rate at which API keys, tokens, and other secrets end up in repositories — a growing concern for security teams.
This trend could place significant strain on development teams and security operations, as AI-generated code becomes more widespread across industries. Organizations adopting AI coding tools may face increased exposure to credential theft, potentially affecting customers whose data relies on secure systems. The findings may also prompt companies to reevaluate their secret-scanning practices and AI governance policies, though the full scope of impact will depend on how quickly mitigation strategies evolve.