Anthropic called for a slowdown in global AI development, saying that its models were increasingly capable of autonomously designing and developing their own successors.
Claude now writes 80% of Anthropic’s code, the firm said, and proposes research directions and solves open-ended problems. AI safety thinkers have long warned of “recursive self-improvement” leading to an exponential increase in AI capabilities, and Anthropic said that such an explosion could be close.
Ensuring powerful AI systems are safe could take time, the company’s cofounder wrote, and “it would be good for the world to have the option to slow or temporarily pause frontier AI development” to give that time.