An artificial intelligence agent blast radius describes the potential reach of an agent failure. It depends on the agent's permissions, credentials, accessible workloads, action rate, delegation paths and whether changes can be reversed.
Teams can reduce blast radius with scoped identity, budgets, rate limits, approval gates, trip wires and recoverable operations. Controls should be enforced by infrastructure because text instructions alone cannot reliably contain a destructive loop.
