Anthropic Updates Responsible Scaling Policy — Flexible Protection System for Growing AI Models

Anthropic published a significant update to its AI risk management policy (Responsible Scaling Policy). Instead of a one-size-fits-all approach, the company introduces a flexible system of proportional protection: ASL safety levels scale together with growing model capabilities. Capability thresholds have been introduced — specific AI abilities that, when reached, require stricter safety measures and monitoring. Current Anthropic models operate at ASL-2 level.