AI Labs Near Breakthrough in Self-Improving Models

AI developers are nearing "recursive self-improvement," where AI models autonomously enhance themselves and build successors, a prospect that promises scientific advances but raises safety concerns about human control. Anthropic reported its Claude model now leads 26% of its research and development, completing tasks end-to-end under human supervision, while OpenAI announced an automated "research intern" and a goal of an AI researcher by March 2028. Some experts warn of runaway superintelligence risks, though others note AI-assisted development has occurred for years. Calls for a coordinated AI slowdown have emerged, with Anthropic open to pausing if competitors verifiably comply, while OpenAI says it values RSI because automated researchers could also advance safety research, though it acknowledges alignment progress may not keep pace