AI Agent Threat Response: Why Pre-Runtime Controls Matter More Than Runtime Detection GitGuardian Blog