Description

A recently disclosed supply chain vulnerability in Anthropic’s Claude Code GitHub Actions integration exposed numerous repositories to potential compromise through a single malicious GitHub issue. Security researcher Ryota K of GMO Flat Security identified several security weaknesses that enabled attackers to circumvent permission controls and inject untrusted input into automated CI/CD pipelines. The primary issue stemmed from an inadequate permission validation process that allowed GitHub Apps to bypass standard write-access verification. Since GitHub Apps can interact with public repositories without requiring installation, threat actors could create malicious applications and submit issues or pull requests capable of triggering vulnerable workflows. After the workflow was triggered, attacker-supplied content was processed without sufficient safeguards, creating an avenue for prompt injection attacks. Researchers demonstrated that specially crafted payloads embedded in GitHub issues could influence Claude Code to perform unintended actions, including accessing sensitive environment variables available during runtime. One of the most severe consequences involved the exposure of OpenID Connect token request credentials used by GitHub Actions for workflow authentication and token generation. By obtaining these credentials, attackers could reproduce the authentication sequence and acquire highly privileged access tokens with write permissions to affected repositories. This level of access could enable unauthorized code modifications, workflow manipulation, pull request tampering, and the introduction of malicious components into downstream software projects. The overall risk was amplified by insecure default configurations frequently found in affected environments. Certain settings allowed external users to initiate workflows, while excessive token privileges and exposed secrets created additional opportunities for privilege escalation. Researchers also demonstrated a multi-stage attack path in which limited access gained through one workflow could later be leveraged to compromise a more trusted process. The vulnerabilities extended to Anthropic’s own repositories, highlighting the broader supply chain implications. Anthropic addressed the issues in Claude Code version 1.0.94 by strengthening validation controls, restricting unsafe workflow triggers, improving command execution safeguards, and reducing opportunities for data exfiltration. Security teams are encouraged to review workflow permissions, limit token scope, and strengthen protections against untrusted inputs.