AI Tooling Fuels Secrets Leakage Across Code, Tools and Developers

Du'aine Davis

Observed in the wild

• In 2025, organisations of all sizes were impacted by a sharp increase in exposed credentials as software development accelerated. GitGuardian identified 28.65 million new secrets committed to public GitHub in 2025, a 34% year-on-year rise, with AI-service credentials increasing 81%. The issue spans public and private repositories, collaboration tools, CI/CD systems, and developer workstations, with long-standing secrets often remaining active into 2026.
• Exposure primarily resulted from hardcoded secrets in source code, AI-assisted commits, MCP and AI infrastructure configuration files, and operational tools such as Slack, Jira and Confluence. AI coding assistants increased development speed but also amplified insecure patterns, particularly where documentation encouraged embedded keys. Supply-chain and prompt-injection-style attacks targeting developer machines and CI runners further enabled large-scale credential theft rather than direct exploitation by a single threat actor.
• Organisations are advised to reduce hardcoded secrets, rotate and retire stale credentials, and treat developer endpoints and CI/CD runners as high-risk assets. Within Microsoft Defender, prioritise alerts for suspicious token use, abnormal service account authentication, and credential access from developer devices. Enforce least privilege, monitor non-human identities, and implement preventive controls such as secret scanning, endpoint protection, and rapid credential revocation workflows.

The State of Secrets Sprawl 2026: AI-Service Leaks Surge 81% and 29M Secrets Hit Public GitHub