Definition
Plain language
How far the damage would spread if an action goes wrong.
As stated in the literature
The scope of systems, data, or users affected by a mistaken or malicious operation; over-broad permissions and control-plane actions increase it, and it appears as an explicit hazard dial in agent-safety benchmarks like UnderSpecBench.
Why it matters: It captures how much damage a single wrong action can cause, so limiting an agent's permissions keeps mistakes from spreading across an entire system.
For example, deleting one customer's record has a tiny blast radius, but wiping the shared database everyone depends on has an enormous one.
Heard on the show
“If a single compromised agent can directly inject into the context of peer agents in a workflow, the blast radius gets much wider.”Episode 030 — Why Your AI Agent Won't Stop Working — and Each Model Falls for a Different Trap