2026-W16 · -1.9
This week’s dossier points to a modest increase in concern, mainly from research on power-seeking evaluations, evaluation awareness, agent security vulnerabilities and faster software-engineering automation. The evidence is mostly laboratory, benchmark-based or forecast-based, not verified real-world loss of control or catastrophic harm. There were no concrete incident or governance documents strong enough to move those categories. The overall index remains in a moderate-risk range and should not be read as a probability of extinction.
A conventional composite index, not a scientific probability of extinction.