AI Coding Tools Expose 29 Million Secrets in 2025

In 2025, over 29 million secrets were leaked due to poor credential management in AI agents, according to GitGuardian's report. Organizations are increasingly integrating AI coding tools like GitHub Copilot, which can inadvertently expose sensitive information during development. The report highlighted a staggering 28,649,024 new secrets found in public GitHub commits, marking a 34% increase from the previous year. This surge in exposed secrets poses significant risks to organizations as developers may unintentionally share API keys and other credentials while using these tools. The issue stems from inadequate security practices surrounding the use of AI in software development. As AI tools become more prevalent, the potential for credential leaks is expected to grow unless organizations implement stricter controls. The current status indicates a pressing need for improved security measures in AI-assisted coding environments.

Key Points: • Over 29 million secrets were leaked in 2025, a 34% increase from 2024. • AI coding tools can expose sensitive information during development processes. • Organizations must adopt stricter security practices to mitigate risks associated with AI tools.