Anchored, not vibes
Each of the 75 questions has written anchors that say what a 10, a 30, a 50, a 70 and a 90 look like, often modelled on a story we already know, from HAL 9000 to Neuromancer's Turing Police.
CARI Live · updated daily
The Civilizational AI Risk Index answers with one number from 0 to 100. It is built from 75 anchored questions in six dimensions and moves each morning with the news. On this scale 2019 sits near 10, and 90 marks the point where humans can no longer correct critical AI systems.
What it is
AI risk is usually argued one headline at a time: a new model, an incident, a law. CARI puts those signals on one scale, with written rules for what each level means, so that a reading can be compared across days and years and checked by anyone.
Each of the 75 questions has written anchors that say what a 10, a 30, a 50, a 70 and a 90 look like, often modelled on a story we already know, from HAL 9000 to Neuromancer's Turing Police.
Every morning the day's AI news is screened and scored against the questions it bears on. Each story can move a dimension only a little, the total is capped, and its effect fades as it ages.
The scorecard answers underneath change only through a recorded review with evidence, so one news cycle cannot rewrite the index.
The same scorecard was answered for points back to 2019, so today's number sits on a long line rather than standing alone.
Each day's full dashboard shows every scorecard answer and every scored story. The daily readings are free to download as CSV or JSON.
CARI is the starting point of MoralWorld, our simulation of how AI could reshape 178 countries.
Six dimensions
The overall index is a weighted average of six dimensions. Each is scored from its own anchored questions. The latest level of each dimension is shown when the live index can be reached.
What the most capable systems can do, how broadly, and how fast that is improving, including AI that speeds up AI research.
How widely that capability spreads: open models, everyday use, critical systems, and the race to have it.
Harm from AI that fails or behaves badly without anyone intending it, from individual cases to systemic shocks.
Whether humans can still see, steer and stop AI systems: deception, resisting shutdown, escaping limits.
The gap between the technology and the rules and institutions able to oversee it, nationally and between states.
Deliberate harm with AI: weapons uplift, cyberattacks, fraud, and the loss of trust in what we see and hear.
Reading the number
Three lines frame every reading, borrowed from how risk is managed in practice. Ten named stages describe the ground between 0 and 100.
Why it matters
CARI is a structured judgement under a stated method, not a measurement and not a forecast. Its value is that the judgement is written down, repeated every day by the same rules, and open to challenge.
A consistent daily series with its inputs attached, to compare against your own view or use in your own work.
A shared language of appetite, tolerance and capacity, and a record of which kinds of events actually move the risk.
A quick, honest answer to "is it getting worse?", with the stories behind the number one click away.
Cite
If you use CARI readings in research, teaching or writing, please cite the index and the date of the reading you used.
CUSP42 Labs (2026). Civilizational AI Risk Index (CARI), daily readings. https://cari.cusp42.com