Agent Hijacking

Reason about agent hijacking as a defender. Learn the conceptual attack family (an autonomous agent gets redirected to attacker goals via injection, environment manipulation, or tool-output crafting), the agent-trust boundary, real-world precedents from frontier-lab disclosures and academic research, and the containment-defence pattern (capability gating, human-in-loop on irreversible actions, scope shrinking, runtime monitors).