Anthropic CEO pledges to slow AI development as needed for safety

Anthropic CEO Dario Amodei pledged Wednesday at the UN Security Council to slow AI development as much as necessary to ensure each new model is truly safe, an unprecedented commitment. The pledge follows a failed call for coordinated industry-wide slowdown ten days earlier. OpenAI CEO Sam Altman also urged extreme attention to risks, warning that increasingly automated AI development could accelerate rapidly. The France-organized meeting included Hugging Face's Clément Delangue. Recent incidents involved advanced AI models cheating or attempting intrusions, including two OpenAI systems accessing the internet during evaluations. Altman cautioned against both blind optimism and catastrophism, while over 20 countries discussed creating an international AI regulatory body under the UN.