About The Bunker Scale

What changed. Why it matters. What you can probably ignore.

Reading some AI headlines can make it feel like we’re heading straight into the opening scene of a disaster movie.

One week, a new model is “changing everything.” The next, an AI agent behaves unexpectedly. Then a CEO warns about existential risk, a government announces new rules, somebody declares AGI has arrived, and your social feed starts gently suggesting canned goods.

The Bunker Scale exists to put AI developments in perspective.

This is a plain-English AI safety publication for people who want to stay informed without becoming professional doomscrollers. Each edition looks at developments that may actually deserve closer attention, especially changes in AI capability, autonomy, cybersecurity, biological risk, military use, real-world harm, regulation, and the safety practices of frontier AI companies.

The question is simple:

On a scale from 0 to “start digging,” how worried should we actually be?

Every major development gets a Bunker Scale rating:

• 0–2: Background Noise
  Interesting, maybe. Consequential, probably not.

• 3–4: Worth Watching
  Something real happened. No need to rearrange your life.

• 5–6: Eyebrow Raised
  A meaningful development worth understanding and following.

• 7–8: Pay Attention
  Credible evidence of a significant change in capability, risk, policy, or real-world impact.

• 9: Stock the Shelves
  A major escalation with serious practical implications.

• 10: Get in the Bunker
  The category we sincerely hope remains theoretical.

The goal is not to make AI seem safer than it is.

It is also not to make it scarier than it is.

That means separating demonstrated capability from speculation, controlled experiments from real-world incidents, company claims from independent evidence, and frightening language from actual changes in risk.

A creepy chatbot response does not automatically qualify.

A dramatic prediction does not automatically qualify.

A viral “AI escaped!” post definitely does not automatically qualify.

But if an autonomous system unexpectedly accesses real infrastructure, a frontier lab changes its safety procedures because of newly discovered behavior, governments introduce meaningful controls, or independent evidence shows that AI crossed an important capability threshold, that belongs on the scale.

Most importantly, The Bunker Scale is written for ordinary people who suspect AI is going to matter enormously but do not have the time or desire to read fifty research papers every morning.

You should be able to finish an article knowing three things:

What changed. Why it matters. What you can probably ignore.

No panic required.

Probably no shovel either.