Inside the AI Supply Chain: Security Lessons from 10,000 Open-Source ML Projects

Analysis of 10,000 open-source AI/ML repositories found that 70% have critical or high-severity vulnerabilities in GitHub Actions workflows — primarily unpinned third-party actions, script/command injection, over-privileged GITHUB_TOKEN, unsafe triggers, and leaked/hard-coded secrets — creating large-scale risk of code injection, credential theft, model/data exfiltration, and repository takeover; the report ranks the top vulnerability classes, provides prevalence metrics, and offers concrete mitigation guidance (pin actions, enforce least-privilege, remove injection vectors, harden triggers, adopt OIDC and improved secret handling).