Blast radius
gobernanzablast radius of an agent action · scope of damage
Blast radius is the maximum damage an AI agent action can cause if it goes wrong: how many customers, systems or euros it reaches, and how costly it is to undo.
Implementa's formulation. Before deciding whether an agent acts alone or asks first, it is not enough to ask whether it is usually right: ask how much it breaks when it is wrong. Blast radius is estimated from three signals: scope (one record or thousands), reversibility (whether it can be undone, and at what cost) and visibility (whether anyone would notice in time). Several IT-operations vendors use it in 2026 as a criterion for granting an agent more autonomy, alongside human approval. For a business with no technical team it is the question that orders what to automate first: small, reversible actions alone, large-radius ones with a person in front.
How it differs from
- Autonomy level
- Autonomy level says how much the agent decides; blast radius measures how much damage each decision can hold and is used to set that level.
- Kill switch
- A kill switch limits how long damage lasts once detected; blast radius limits its size before the agent acts.
Examples
- An agent that deletes a draft has a small radius; one that emails an invoice to the whole customer base has a large radius and cannot be undone.
- An agent that changes a product price on the website affects every sale until someone notices.
FAQ
- How do I estimate the blast radius of an action?
- Score three things: how many records or people it reaches, whether and at what cost it can be undone, and who would notice if it went wrong. The wider, harder to undo and less visible, the larger the radius.
- Does a small radius mean I can drop supervision?
- It lets you reduce it, not remove it. Replace prior approval with after-the-fact sampling and a log of what ran.