AI Extinction Clock ยท Severity Scale

Reference

Severity Scale

Levels 1–10 · adopted in v3.0

What “catastrophe” and “extinction” mean on this site, defined on one scale so every reading says what it is a probability of.

Ten levels of harm

The scale runs from a single incident to the end of all life on Earth. The Editorial Opinion clock measures the chance of an event at level 5 or worse, shaded below, within a rolling five years.

LevelNameThreshold
10Extinction of lifeAll life on Earth ends
9Human extinctionNo humans survive
8Civilizational collapseMost of humanity dies; recovery uncertain
7Loss of controlHumanity permanently loses the ability to direct or correct AI, even without mass death
6Global catastrophe10 million or more deaths, roughly pandemic scale
5Multinational catastrophe100,000 or more deaths, or a global financial or infrastructure crisis
4National catastropheThousands of deaths; a nation's critical systems compromised
3DisasterMore than 50 deaths or $1 billion in damage
2Serious incidentMajor breach or infrastructure disruption; first deaths
1IncidentIndividual harm, fraud, data exposure

Level 3 matches the threshold in Anthropic's proposed federal bill, which would let a court pause a model posing an imminent risk of more than 50 deaths or $1 billion in damage.

Why loss of control is level 7

Many researchers count a permanent loss of human control over AI as an existential outcome even if people survive. We place it at level 7, above global catastrophes measured in deaths and below the collapse of civilization. That placement is an editorial choice and is open to revision.

How the scale is used

  • Every item in Latest Developments will be tagged with the level of harm it involved or points toward.
  • Expert estimates will be recorded with the level and horizon each expert meant, where they say.
  • Those records feed the published trends that the audited review weighs before setting either clock.