Recursive self-improvement (RSI) — the theoretical ability of an AI system to autonomously upgrade its own intelligence in a runaway cycle — has emerged as the defining frontier of the global AI race, with leading companies in both the United States and China accelerating efforts to reach it first. The competition intensified in the week of 12 June 2026, after Claude developer Anthropic announced it was edging closer to RSI following recent model advances, including the public release of its Mythos model on Tuesday.

What is recursive self-improvement and why it matters

The concept of RSI traces back 61 years to British mathematician Jack Good, who coined the term 'intelligence explosion' to describe a machine entering a fully automated self-improvement loop that would rapidly surpass human intelligence. The strategic logic is stark: the first country or company to achieve RSI would cement an effectively unassailable lead over every competitor. That framing has made RSI a declared priority for top-tier AI labs in the US.

Anthropic edges closer, calls for global pause option

Anthropic went viral last week after publicly stating it was approaching RSI, attributing the progress to advances exemplified by its Mythos model. The company simultaneously called for 'the option' to temporarily pause AI development globally, citing heightened risks of losing control of AI systems once RSI is achieved. Critics have characterised such warnings as marketing hype rather than genuine safety advocacy.

China's competitive push: self-evolution on the agenda

Chinese AI companies and researchers are no longer treating RSI as a distant abstraction. At the Zhongguancun Forum — China's premier state-backed technology event — Luo Fuli, lead developer of Xiaomi's MiMo AI model, told an audience of major policymakers and investors in March that 'self-evolution' would be the biggest trend in AI over the next year. 'We've already begun to see an implementable and actionable path to achieve AI model self-evolution,' she said.

The competitive backdrop: key players and technologies

The RSI contest spans a wide field of organisations. On the US side, Anthropic is among the most vocal, but the race also involves firms relying on Nvidia CUDA-based infrastructure to power increasingly autonomous AI research pipelines. In China, companies including Alibaba Group Holding — developer of the Qwen model family — ByteDance, and MiniMax are all active in the self-improving AI space, alongside the ASI-Evolve research initiative and Xiaomi's MiMo.

What's next

The convergence of public announcements from both US and Chinese labs signals that RSI is transitioning from theoretical milestone to near-term engineering target. Observers will be watching whether Anthropic's call for a global development pause gains traction among regulators, and whether Chinese state-backed forums translate Luo Fuli's roadmap into coordinated national funding. The company or nation that demonstrates a credible, controlled RSI capability first will face immediate pressure — from rivals, regulators, and the public — to define the rules of what comes next.