Claude Code and Gemini CLI Flaws Let a GitHub Issue Reach CI Workflow Secrets
Summary
Researchers discovered vulnerabilities in Anthropic's Claude Code and Google's Gemini CLI that allowed an attacker to execute code on CI runners. The flaws enabled code execution even from GitHub issues created by accounts without repository privileges, and on OpenAI's platform, it led to hijacking the next agent run.
IFF Assessment
These vulnerabilities represent a significant security risk, allowing unauthorized code execution and hijacking of agent runs, which are detrimental to defenders.
Severity
The vulnerabilities allow for unauthorized code execution within CI/CD pipelines, which typically have access to sensitive secrets and infrastructure. The attack vector is relatively straightforward via GitHub issues, and the impact includes potential data exfiltration, system compromise, and service disruption.
Defender Context
This incident highlights the critical need for secure configuration of AI coding agents and their associated CI/CD pipelines. Defenders should scrutinize access controls and secret management for these tools, as compromised agents can lead to significant downstream security incidents.