What frameworks do you use to assess the effectiveness of safety programs in Character.AI-like products?
