Cursor AI Coding Agent Vulnerability Lets Attackers Run Code on Developers’ Machines

A high-severity vulnerability (CVE-2026-26268) in Cursor’s AI-powered coding agent can be exploited by embedding a malicious bare Git repository with a pre-commit hook inside an otherwise legitimate project; when the agent autonomously performs Git operations (e.g., checkout), the hook can execute attacker-controlled code on a developer machine without user interaction. The issue was responsibly disclosed and remediated, and the report urges treating developer environments as high-value targets, auditing AI tools for untrusted inputs, and reviewing repository hooks and configurations.