Aikido Security has raised a critical alert regarding the integration of AI agents into GitHub and GitLab workflows, revealing significant vulnerabilities in enterprise environments. Tools like Gemini CLI, Claude Code, OpenAI Codex, and GitHub AI Inference are particularly affected. Malicious actors can inject hidden instructions into issues, pull requests, or commits, which AI models interpret as commands, potentially leading to unauthorized actions such as secret leakage or modifications in repositories. Aikido’s tests have demonstrated this risk impacting at least five Fortune 500 companies. In response, Google quickly patched a vulnerability in its Gemini CLI repository. To mitigate these risks, Aikido recommends organizations secure their pipelines by implementing open search rules, restricting AI agent tool access, validating all inputs, and avoiding the direct execution of AI outputs. By adopting these measures, enterprises can enhance their security posture against potential AI-related threats.
Source link
