Anthropic co-founder Jack Clark has moved up his forecast for when AI systems could fully improve themselves. He previously expected full recursive self-improvement (RSI) in late 2028. He now expects it in 2027.
Clark made the update on October 12, 2026, in his Import AI newsletter. His reason was not a single breakthrough. It was a run of benchmark data showing rapid gains in areas like mathematics and automated research and development.
What recursive self-improvement actually means
In technical terms, full RSI describes autonomous AI systems that can enhance their own design and capabilities without human intervention. Put simply, the model stops waiting for people to build the next version and starts building it itself.
The numbers behind the shorter forecast
In August 2026, AI agents built on the company’s Claude model led 26% of core AI R&D tasks.
Over 90% of the tasks the agents took part in reached at least a collaborative level of involvement, classified as AL3 or higher.
Approximately 30,000 AI agents are running concurrently at Anthropic, according to the research findings. Safety monitors overseeing those agents reportedly show low override rates.
Vals AI’s RSI Index predicts that full automation of machine learning engineering work at Anthropic could arrive around July to August 2027.
A CEO who has been urging caution
In September 2026, Anthropic CEO Dario Amodei published a post calling for frontier AI development to be paced, acknowledging that the risks tied to RSI are accelerating. Amodei has advocated for coordinated industry efforts to manage those risks.
What this means for the AI industry
For policymakers, the shortened timeline compresses the window for action. Rules designed around a late-2028 arrival look different when the expected date shifts to 2027. Amodei’s call for coordinated pacing is, in effect, an invitation for governments and industry to move before the forecasts become the present.
There are a few concrete things to watch from here. The first is whether the 26% figure for agent-led R&D tasks keeps climbing in Anthropic’s future disclosures. The second is the Vals AI RSI Index. Its July to August 2027 estimate for full ML engineering automation at Anthropic is a dated, checkable prediction. The third is how those override rates evolve as the agent count grows. Approximately 30,000 concurrent agents with little human correction is a remarkable operational fact.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.