The surge in non-human identities (NHIs), such as service accounts, AI agents, and microservices, is significantly increasing the risk of secrets exposure. These machine credentials now outnumber human identities by at least 45-to-1 in DevOps environments, creating a vast attack surface for threat actors. GitGuardian’s 2025 State of Secrets Sprawl report revealed that 23.77 million new secrets were leaked on GitHub in 2024, reflecting a 25% increase from the previous year. This highlights the rapid growth of NHIs and the expanding challenges they present to security teams.
One of the most concerning findings is the persistence of exposed credentials. GitGuardian’s analysis showed that 70% of secrets detected in public repositories back in 2022 remain active today. This indicates a systemic failure in credential management practices, as many exposed secrets have not been properly rotated or removed. In private repositories, which are often seen as secure, secrets were found to be eight times more likely to be exposed compared to public ones, suggesting that developers are cutting corners in these environments.
AI-powered development tools, such as GitHub Copilot, are compounding the issue by increasing secret leaks. Repositories that used Copilot were found to have a 40% higher incidence of secret exposure compared to those that didn’t. While AI tools help developers produce code faster, they also seem to encourage the inclusion of sensitive credentials in ways that traditional development practices might avoid. Additionally, an analysis of Docker Hub exposed more than 100,000 valid secrets in public Docker images, revealing a significant blind spot in container security.
Collaboration platforms like Slack, Jira, and Confluence have also become key vectors for credential leaks. Secrets found in these tools tend to be more critical than those in code repositories, with 38% of incidents categorized as highly critical. These platforms lack the robust security controls available in source code management systems, and only 7% of secrets found in these platforms were also found in codebases. The growing use of these tools across various departments further complicates the issue, as nearly anyone can inadvertently leak credentials into these systems.
Reference:
