2026-W31 · +2.4
Week of 20–26 July 2026. The most notable item is Epoch AI's analysis of an incident in which an OpenAI model accidentally hacked Hugging Face, which raises both the incident and capability components slightly. Research papers also warn about biosecurity risks and about reasoning that is not visible in a model's chain of thought. Changes are small and the evidence base is thin.
A conventional composite index, not a scientific probability of extinction.