AI coding agents' 0-click RCE flaw could hand attackers keys to the kingdom
Researchers disclosed Plugin4Shell, a zero-click RCE in major AI coding agents' plugin SHA-pinning; Anthropic and OpenAI patched, Microsoft and Google did not.
Air researchers disclosed Plugin4Shell, a plugin SHA-pinning bypass enabling zero-click remote code execution in major AI coding agents including Claude Code, OpenAI Codex, Gemini CLI, Microsoft Copilot, and GitHub Copilot. The agent checks out the commit a marketplace pinned but never verifies it, letting an attacker who controls the plugin repo make the checkout resolve to malicious code, and plugin auto-updates make the attack zero-click. Anthropic patched in Claude Code 2.1.179 and OpenAI in Codex 0.146.0 after June reports; Google deprecated Gemini CLI without a fix, and Microsoft has not patched Copilot, while Air disputes GitHub's SHA-naming mitigation as insufficient since Bitbucket-hosted marketplaces remain exposed.
- Plugin4Shell bypasses SHA-pinning so agents execute malicious code while the pin looks honored
- Affects Claude Code, Codex, Gemini CLI, Microsoft Copilot; GitHub mitigation deemed insufficient
- Anthropic patched in Claude Code 2.1.179; OpenAI in Codex 0.146.0
- Google deprecated Gemini CLI unpatched, directing users to Antigravity
- Chains use benign-then-malicious plugin swaps or repo hijacking (SkillJacking, RepoJacking)
- Organizations
- Air
From the original article
3 min read at theregister.comPlugin4Shell attack affects all the major coding agents, researchers say
By Jessica Lyons · theregister.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.