Anthropic CEO Dario Amodei urged leading AI companies to slow work on their most advanced models and allow independent safety reviews, saying more capable systems could pose growing cybersecurity and self-improvement risks.

OpenAI CEO Sam Altman endorsed Amodei’s proposal for independent evaluators with employee-like access, and Elon Musk wrote that “Dario is right.”

Anthropic said it will give third-party evaluators broad access to verify safety practices and report incidents, including placing embedded evaluators inside its offices.

Separately, a departing Anthropic researcher warned about AI dangers and said the chance of an extreme catastrophic outcome was more than 10%.

Broader coordination could be difficult because an industry-wide slowdown may raise antitrust issues.

🔗 Source: Bloomberg

Recent Anthropic developments