“Anthropic's Claude 4 Opus model is classified as AI Safety Level 3 (ASL3) because it provides a significant uplift in accessing information for biological harm compared to a Google search.”