Comment by sanxiyn

Agreed, and that serves Anthropic. It seems unproblematic to me. Dario probably sincerely believes in mandatory safety testing for capable models (open and closed), and likes the fact that it aligns with Anthropic's interest.