> Prompt injection to RCE in AI agents

[SOURCE] Trail of Bits [DATE: 22/10/2025 11:00] [LANGUAGE: EN]

We bypassed human approval protections for system command execution in AI agents, achieving RCE in three agent platforms.