A forecast says a “superhuman AI researcher” that surpasses the world’s top human researchers is likely to appear in 2027. The analysis says AI is speeding up AI research and development, pushing technological progress faster than previously expected.

Gigazine in Japan reported on Nov. 11 that the nonprofit research group AI Futures Project, which includes former OpenAI governance researcher Daniel Kokotajlo (대니얼 코코타일로), recently released a forecast bringing forward the expected arrival of a superhuman AI researcher to 2027.

The change was prompted by findings from METR, a U.S. nonprofit AI evaluation organization. METR on Nov. 5 released a report analyzing how much AI contributes to Anthropic’s AI research and development.

METR examined Anthropic’s internal research materials, model performance and employee interviews from Aug. 12 to Oct. 29. It assessed that as of October, the rate of improvement in AI models’ ability to perform specific tasks was about twice as fast as in March 2024 to October 2025.

In internal benchmarks, the pace of model performance improvement rose by a factor of 1.9. METR said much of that acceleration resulted from using AI in research and development.

Still, AI has not fully automated research and development. The researchers said AI is at the stage of supporting human researchers, but autonomous research capabilities could expand rapidly in the future.

Based on that, the AI Futures Project said a superhuman AI researcher is likely to emerge in 2027. It defined this as AI that understands the research goals of top-tier human researchers, fully automates coding and can conduct R&D.

Kokotajlo previously put the probability of such AI appearing before 2029 at 50 percent, but moved the expected timing forward after reflecting recent research results.

The research group also estimated a 35 to 65 percent chance that AI capable of automating most office work could appear before 2028.

If such technology becomes reality, R&D productivity could rise sharply, but risks could also grow, including large-scale job losses, weaker consumption and failures to control AI.

The researchers said if the emergence of superhuman AI without sufficient safety assurances accelerates more than expected, the pace of cutting-edge AI development should be slowed and an international safety management framework should be put in place.

Still, the forecast is based on limited research data and predictive models, and the actual timing has not been proven.