Anthropic, OpenAI flag recursive self-improvement risk as AI nears RSI threshold

Researchers at Anthropic and OpenAI warn that recursive self-improvement (RSI)—AI building better versions of itself without human guidance—is moving closer to reality, a scenario safety experts have long feared. The claim is based on statements from these labs, not yet on a demonstrated RSI event, so the timeline and severity remain uncertain. This matters because RSI could produce increasingly capable systems that outpace human control, a core existential-risk concern for frontier AI development.