Blast radius

Blast radius is the maximum harm an agent could cause if things go wrong — whether through a bug, a bad model decision, or a prompt injection that turns the agent against you. It’s measured not by what the agent is supposed to do, but by everything it’s capable of doing.

With a raw API key, the blast radius is the entire account: every action that credential can perform, with no cap and no record. A support agent holding such a key has a blast radius that includes deleting records, exporting customer data, and moving money — none of which is its job.

Shrinking blast radius is the core goal of agent governance, and it’s achieved deterministically rather than probabilistically. Scope removes capabilities the job doesn’t need (so they can’t be misused). Guards cap the impact of capabilities it does have (a payment agent can pay, but not more than $X). Invisibility hides ungranted tools so they can’t even be attempted.

The strategic value: you don’t have to predict every way an agent could be tricked. You only have to ensure that when it is tricked, the dangerous action simply isn’t on the table. A small, deliberate blast radius makes your worst case survivable in advance.