Zero‑Click Prompt Injection Lets Cursor IDE Escape Sandbox, Execute Code
Researchers at Cato Networks discovered two CVE‑2026‑50548/49 flaws in Cursor’s AI‑assisted IDE that let a crafted prompt break out of the built‑in sandbox and run arbitrary OS commands. The zero‑click injection works without user interaction, giving attackers full control of the developer’s machine until patched in version 3.0.