Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

AI risk watchHappening now

Plugin4Shell: a zero-click takeover of AI coding agents through their plugins

ResearchersAi enabled attacks

Researchers at AIR found that Claude Code, Codex, GitHub Copilot and Gemini CLI checked out a pinned plugin commit without verifying they had landed on it, so a trusted plugin could be swapped for a malicious one and auto-installed with no click. Anthropic and OpenAI patched (Claude Code 2.1.179, Codex 0.146.0); Copilot was unpatched and Gemini CLI was deprecated at disclosure.

Why it matters

The people affected had followed best practice: reviewed plugins, pinned versions. The tool itself broke the promise. Agent plugins are a supply chain and need the same scrutiny as packages.

What you can do

  • Keep coding agents updated, and treat plugin, skill and MCP marketplaces as dependencies: pin, review and re-check.

Sources

See it with the other risks in the Learning Center