Anthropic proposes a structured, verifiable pause on frontier AI development to address risks of recursive self-improvement.
