Agent safety is the engineering of bounded autonomy — the five-layer containment stack: bounded permissions (what the agent cannot do, enforced by con… First-principles module in Autonomous Systems, with a mastery quiz.