Safety2h ago
Safety
Anthropic Formalizes Three Catastrophic Risk Models
A 186-page report from Anthropic introduces three distinct threat models for catastrophic AI risk. The company specifically formalizes misalignment in high-stakes settings while addressing automated R&D and bio-weapon production. This framework provides a structured approach to safety auditing. Practitioners can now map specific technical failures to these defined catastrophic outcomes.