China researchers' "Last AI" roadmap draws RSI ladder

The paper's ladder is not a claim that AI has reached self-improvement; it is a scoring scheme. The authors placed 491 existing papers on it, and the distribution they report is bottom-heavy: 75.4% at Levels 1 and 2, where humans still design the upgrades or the AI only proposes fixes, against 5.9% at Level 5, where AI redesigns the improvement process. They also note that sustained gains across successive model generations remain an open problem, so the upper rungs describe a direction rather than a demonstrated capability.

The roadmap arrives alongside lab policies that approach the same territory from a different angle. OpenAI's Preparedness Framework puts fully automated AI research and development at its Critical threshold, with one indicator being sustained improvements across model generations taking one-fifth as long as equivalent work in 2024, and specifies halting further development until safeguards are defined. Anthropic's Responsible Scaling Policy, effective July 8, sets a threshold around replacing its entire research scientist and research engineer workforce at competitive costs, and the company acknowledges the threshold has been hard to apply. Google DeepMind's Frontier Safety Framework sets a threshold at fully automating a Google team focused on improving AI capabilities, at roughly comparable total costs.

Those frameworks measure speed, scale and risk, while the paper's ladder measures control over the improvement process, and the body notes that a Level 5 label alone therefore gives an incomplete picture of the safety question. The comparison also has a national dimension with limits: the paper reflects its authors' outlook, and AP reported that Guo Jiakun, a spokesperson for China's foreign ministry, rejected AI warnings aimed at China while calling for cooperation on global AI governance, suggesting resistance to the geopolitical framing. For labs and safety teams, the practical test is likely to be narrower than any label: how much research a system completes, how long it takes, what it costs and how much human help it needs, and who inside a lab can check changes and block a release.