2026-W18 · +0.6
This week, research highlighted methods for AI systems to discover safety objectives and measure propensity for unsanctioned behavior, indicating notable risks of autonomy. Advancements in automated red teaming and adversarial data generation suggest progress in dangerous capabilities. No significant new incidents were reported, but governance challenges were noted in a draft AI policy. Claims of new models closing performance gaps and the veto of a data center moratorium suggest fast but not explosive progress in AI development.
A conventional composite index, not a scientific probability of extinction.