AI labs near breakthrough in self-improving models

Leading AI developers say artificial intelligence models are nearing the ability to improve themselves autonomously, a concept known as recursive self-improvement, which could accelerate advances in science but also raises fears about AI evading human control. Anthropic reported this week that its Claude model now leads 26% of the company's model research and development, completing most tasks end-to-end under human supervision, while OpenAI has developed an automated "research intern" and aims for a fully automated AI researcher by March 2028. Some executives, like Elon Musk, predict full autonomy in model improvement could arrive by the end of this year or by 2027 The push for RSI has sparked industry divisions, with calls for a coordinated slowdown to ensure safety measures keep pace with capabilities. Anthropic has said it would pause development if competitors verifiably did the same, while OpenAI acknowledged it does not yet know how to safely achieve aligned, full RSI, though it values the goal because automated researchers could also aid in safety and alignment work