1 of 1
Story summary
- Dario Amodei, Anthropic CEO, proposed embedding third-party safety evaluators within the company’s AI development.
- The plan gives evaluators internal-risk-team access and permits publishing findings with limited redactions.
- Julie Andersen Hill, dean of Wyoming College of Law, said evaluators lack a “kill-switch” authority.
- Deborah Raji, UC Berkeley researcher, warned access alone does not guarantee evaluator independence or audit standards.