AI labs say self-improving models are close

Leading AI labs say artificial intelligence models are nearing the ability to improve themselves autonomously, a milestone called recursive self-improvement that could accelerate scientific advances but also raises fears about AI evading human control. Anthropic reported its Claude model now handles 26% of its research and development tasks under human supervision, while OpenAI aims to create an automated AI researcher by March 2028. Some executives, including Elon Musk, predict fully autonomous model improvement could arrive by 2027. Experts note AI has assisted in model development for years, but the shift toward full autonomy has sparked industry divisions over safety, with some calling for a coordinated slowdown while others pursue the technology aggressively.