Defense
AI agent red teaming
Adversarial testing of AI agents before production — simulating prompt injection, tool abuse, goal hijacking, jailbreaks, and multi-step attack chains. Targets the agent's reasoning and decision-making, not its network exposure. Distinct from traditional penetration testing.