Malicious Input

Google Patches Antigravity Flaw That Let Prompt Injection Break the Sandbox

Researchers say weak checks on native tool inputs expose AI coding agents to code execution.