CARICivilizational AI Risk Index

How CARI works

CARI is a single number for where AI risk stands. It runs from 0 to 100. The scale is anchored so that 2019 sits near 10, and 90 means humans can no longer correct critical AI systems.

A scorecard of anchored questions

The base comes from 75 questions, each with written anchors that say what a 10, a 50 or a 90 looks like. The questions sit in six dimensions, combined with fixed weights:

  • Capability (weight 20%)
  • Diffusion (weight 15%)
  • Accident (weight 18%)
  • Control (weight 20%)
  • Governance (weight 15%)
  • Misuse (weight 12%)

Answers change only through a reviewed, recorded decision with evidence. The same scorecard was answered for mid-year points back to 2019, which gives the long history.

A daily news adjustment

Each morning the day's AI news is collected, screened for relevance and scored against the questions it bears on. Each story can move a dimension only a little, the total adjustment per dimension is capped, and the effect of a story fades as it ages. So the daily number reacts to events without letting one news cycle rewrite the scorecard.

Lines and stages

Three lines frame the reading: appetite at 40 (above it, monitor and report), tolerance at 60 (above it, escalate) and capacity at 80 (above it, intervention is no longer credible). Ten named stages, from Dormant to Skynet, describe the ground between them.

What CARI is not

It is a structured judgement, not a measurement and not a forecast. The scorecard and the scoring of news involve AI-assisted judgement and can be wrong. The full dashboard for each day shows the scorecard answers and every scored story, so each reading can be checked.

CARI is the starting point of MoralWorld, CUSP42 Labs' simulation of how AI could reshape 178 countries.