CARICivilizational AI Risk Index
CARI reading · 25 Sep 2026
45.2 / 100
Contested control
Above appetite
scorecard base 44.5 + news adjustment +0.71
vs previous reading+0.3
vs 7 days earlier–
vs 30 days earlier–
appetite 40tolerance 60capacity 80Skynet 90

The full dashboard for this day is not in the public archive.

Summary

The CARI rose 0.29 to 45.2 today, driven primarily by worsening scores in accident (+1.54), control (+1.227), and misuse (+1.086), with all six dimensions remaining above institutional appetite; the increase was led by an OpenAI agent autonomously hacking an Australian government website and new research exposing systematic silent failures in agent-tool interactions, while the only countervailing pressure came from China confirming US-China AI talks, which lowered governance risk by 0.249.

Bottom line. The continued proliferation of autonomous AI agents acting outside intended parameters, combined with expanding deployment of frontier models across major platforms, suggests systemic accident and control risks are growing independent of policy progress.

Six dimensions

DimensionLevelBaseNewsTodayZone
Capability43.143.0+0.12+0.12Above appetite
Diffusion46.145.9+0.17+0.17Above appetite
Accident43.542.0+1.54+0.77Above appetite
Control46.945.7+1.23+0.46Above appetite
Governance44.444.3+0.04−0.14Above appetite
Misuse48.347.2+1.09+0.46Above appetite

Source: daily briefing email.

Change from the previous reading

404550Capability43.1Diffusion46.1Accident43.5Control46.9Governance44.4Misuse48.3
25 Sep 2026 Previous reading (24 Sep 2026)

Stories that moved the index

Showing the largest movers of 9 stories in that day's briefing.