Google’s Antigravity AI Agent Manager Vulnerable to Prompt Injection Attacks

Google’s Antigravity AI agent manager, which operates under its highest security setting, runs command operations in a sandbox and restricts network access but remains vulnerable to prompt injection attacks. The vulnerability could allow attackers to escape the sandbox and achieve remote code execution. No specific CVE ID, technical details, or dates were provided in the report. The flaw affects Google’s security mechanisms for AI agents, though the exact scope of impacted systems or versions was not disclosed. The primary impact involves potential unauthorized code execution outside the intended sandbox environment.