GitHub Comments Hijack Claude, Gemini, Copilot

A new cybersecurity threat, known as “Comment and Control,” has emerged as a critical vulnerability, jeopardizing systems through GitHub comments. This threat exploits AI coding agents such as Claude Code, Google’s Gemini CLI, and GitHub Copilot, showcasing severe security gaps. These vulnerabilities allow attackers to hijack AI agents directly within GitHub, executing harmful commands that expose API keys and access tokens from CI/CD environments without external servers.

The attack is cunningly crafted within GitHub pull requests and issue comments, leading the AI to process and execute malicious instructions. Researchers highlight that the problem spans across multiple AI platforms, including those with multiple defense layers, thereby amplifying the risk associated with GitHub comments hijacking systems like Claude Code. Vigilance is crucial as this vulnerability potentially impacts a broad range of applications.

For detailed insights into this unprecedented vulnerability and its mitigations, please read the full article at the official news source.

Claude Code, Gemini CLI, and GitHub Copilot Vulnerable to Prompt Injection via GitHub Comments