Anthropic's HAWK-110 model to surpass 90% alignment score by 2026 Q4

Anthropic's research suggests their latest models (e.g., HAWK-110) have near-zero blackmail attempts in tests. If competitors achieve similar results, Anthropic's alignment claims will be validated, boosting its ethical AI leadership narrative.

Implied probability (Yes)

70%