AI companies in the US and China are increasingly using AI to write code, design experiments and develop training techniques for more capable models, according to the South China Morning Post. The long-term ambition is recursive self-improvement: systems that can help train and build their successors in an accelerating loop.
Watch Desk analysis
What happened
The report says both sides now view recursive self-improvement as a potential turning point in the AI race. OpenAI has described ambitions for an automated AI researcher, while also acknowledging that it does not yet know how to reach fully aligned recursive self-improvement safely.
That distinction matters. The evidence describes a direction of travel and an industry goal, not a demonstrated machine that can independently improve itself without human researchers.
Why it matters
If AI can reliably contribute to the work of building better AI, progress could speed up while becoming harder for humans to inspect. The practical question is not simply whether models can write code today, but whether their contributions can be tested, reproduced and constrained before they shape the next generation.
For readers, the useful dividing line is between AI-assisted research, which is already being pursued, and full autonomous self-improvement, which the supplied evidence does not establish has arrived.
Our read
This is a meaningful shift in what the frontier labs are trying to automate. Treat the grandest claims as a roadmap and warning sign, not a finished capability. The next credible milestone will be a reproducible account of what an AI researcher can improve, under what supervision and with what safety checks.
What to watch
- Whether labs publish concrete examples of AI-designed improvements that humans can independently reproduce.
- How much of the research process remains under human selection and approval.
- Whether safety evaluations keep pace with systems contributing to their own successors.
- Whether US and Chinese labs disclose comparable evidence rather than competing through slogans.
Discussion spark: What evidence would convince you that an AI system is genuinely improving the next generation, rather than simply assisting researchers more efficiently?
Sources and evidence
- The US and China are racing to build ‘self-improving AI’. Here’s what’s at stake – South China Morning Post (13 September 2026, 00:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.