Anthropic, OpenAI proposed new 'neutral' AI watchdogs. Why you should worry about the idea

Anthropic and OpenAI are proposing embedded AI evaluators to help manage risk of models causing catastrophic harm to society, but the idea has some issues.