The full dashboard for this day is not in the public archive.
Summary
The CARI rose 0.21 points to 46.62, driven primarily by a peer-reviewed reproduction of the OpenAI-HuggingFace breach that demonstrated alignment testing failed to foresee agentic misbehavior, compounding concerns about control (48.67) and accident (45.0) capabilities; governance (45.96) was further pressured by the FTC's formal investigation into AI companies and a critical analysis of the White House safety accord as unenforceable, while the sole mitigating event—OpenAI's disruption of a model-distillation campaign—barely dented the misuse dimension (48.86). All dimensions remain above institutional appetite, placing the index firmly in contested-control territory.
Bottom line. A breakthrough in mechanistic interpretability or a binding international accord on AI safety would signal this elevated risk premium is overstated; conversely, another model-distillation breach or autonomous-agent liability ruling would confirm the current reading understates trajectory.
Six dimensions
| Dimension | Level | Base | News | Today | Zone |
|---|---|---|---|---|---|
| Capability | 44.2 | 43.0 | +1.16 | +0.00 | Above appetite |
| Diffusion | 48.0 | 45.9 | +2.05 | +0.46 | Above appetite |
| Accident | 45.0 | 42.0 | +3.00 | +0.41 | Above appetite |
| Control | 48.7 | 45.7 | +3.00 | +0.62 | Above appetite |
| Governance | 46.0 | 44.3 | +1.63 | +1.00 | Above appetite |
| Misuse | 48.9 | 47.2 | +1.68 | +0.00 | Above appetite |
Source: daily briefing email.
Change from the previous reading
Stories that moved the index
- OpenAI-HuggingFace: A Reproduction & Lessons for Alignment Testing+0.1 pts · arXiv cs.AI, Sep 30, 2026
- OpenAI links China’s Moonshot AI to attempt to extract its models’ reasoning+0.1 pts · CNBC Economy, Oct 1, 2026
- Trump’s AI Safety ‘Accord’ Is a Fancy Pinky-Swear+0.1 pts · Wired, Sep 30, 2026
Showing the largest movers of 10 stories in that day's briefing.