Blast Radius

The term originates from blasting technology, where it refers to the area an explosion destroys. In technology, this image has been adopted. The question it addresses is: if one part of a system breaks, what does it take down with it? If a single server fails and no one notices, the radius is small. If the same server fails and logins stop working worldwide, it’s enormous. So the term doesn’t describe a technology, but a property: how strongly a single fault affects the outside world.

Why a single fault can take down entire services

Modern online services consist of hundreds of small programs that call on each other. These building blocks often depend on shared foundations: a database, a network service, a central login system. These very shared points are dangerous. They’re convenient because you only need to build them once. They’re risky because their failure hits everything built on top of them.

The major outages of recent years have almost always followed this pattern. A faulty configuration update gets rolled out, and suddenly thousands of websites become unreachable. Not because all these websites were broken, but because they use the same provider for one small building block. The actual fault was tiny, the blast radius was global.

With AI systems, there’s an additional concern. A language model that automatically sends emails or deletes files can not only display errors but execute them. That’s why security professionals now also talk about the blast radius when it comes to AI tools: what’s the worst this program could do if it makes a mistake or gets manipulated?

How to keep the radius small

The most important tool is isolation. A system is divided into segments that can run independently of one another. Cloud providers do this through regions: data centers in Frankfurt and in Ireland share as little as possible. A fire in Frankfurt shouldn’t affect Ireland. The model here is the bulkheads in a ship’s hull, which prevent a leak from flooding the entire hull.

A second tool is the cautious rollout. New software is first released to only one percent of users, then ten percent, then everyone. If errors show up during this process, they only affect that one percent. This also includes the ability to quickly roll back a change before it has spread everywhere.

Third, one limits permissions. A program is only given access to what it truly needs. A service that resizes images doesn’t need to be able to read customer data. If it gets hacked, the loot is small. The same principle applies to AI assistants: giving a model only read access limits the damage if it does something foolish.

The term in outage reports and security disclosures

After major outages, companies like Amazon, Microsoft, or Cloudflare publish reports that review the incident. The term almost always appears in them. A typical phrasing: measures have been taken to reduce the blast radius of future incidents. Translated, that means: we can’t prevent errors, but we want fewer customers to be affected by them.

The metric is also of interest to investors and customers. When many companies use the same cloud provider, a shared risk emerges. A single outage can then hit banks, online shops, and government agencies simultaneously. Regulators are now closely watching this dependency, since it affects entire sectors of the economy.

A common misunderstanding: blast radius does not mean probability of occurrence. The term says nothing about how often something goes wrong, only how far the damage reaches. A very rare event can have an enormous radius. Only both together produce the risk that security teams plan around.