AI for Science: AlphaFold, Materials, Weather, Math — 2026-09-12

Top developments

OpenAI Claims Solution to Navier-Stokes Millennium Problem

On September 8, 2026, OpenAI announced it had used artificial intelligence to make significant progress on the Navier-Stokes existence and smoothness problem, one of the Clay Mathematics Institute's seven "Millennium Problems." The company describes this as a fundamental transformation of higher mathematics, with Nature reporting that the AI has "cracked the physics of fluids" using advanced modeling techniques. While the full proof has not yet been peer-reviewed by the mathematical community, the announcement has triggered a "scientific arms race" regarding credit and methodology between AI labs and traditional mathematicians.

Google WeatherNext 3 Update Improves Cyclone Forecasting

Google DeepMind released an update to its WeatherNext 3 AI weather model on September 9, incorporating a significantly expanded set of raw satellite data inputs to improve forecast accuracy. The update allows the model to deliver three-day cyclone forecasts with accuracy comparable to previous two-day predictions from traditional numerical weather prediction (NWP) models. This enhancement is now integrated into Google Search, Maps, and Gemini, providing users with more reliable early warnings for severe weather events.

Insilico’s AI Drug Candidate Shows Age-Reversal Potential

Insilico Medicine published results in Nature Biotechnology on September 7 indicating that Rentosertib, an IPF candidate discovered and designed entirely by AI, shows potential for reversing biological age markers. The study utilized six proteomic aging clocks to assess the drug's impact, marking the first time an AI-generated drug for an AI-discovered target has demonstrated such physiological effects in clinical trials. This finding highlights the growing capability of generative AI not just to identify targets, but to design molecules with complex systemic biological impacts.

NetraAI Enhances Clinical Trial Signal Detection

A new peer-reviewed study published in MDPI’s AI Journal on September 10 demonstrated that NetraAI, a specialized AI tool, successfully uncovered clinical trial signals missed by conventional algorithms and foundation models. The study showed that NetraAI’s biological signatures boosted the accuracy of eight evaluated algorithms across every dataset tested, including foundation models like Fable 5. This suggests that niche, domain-specific AI tools may still outperform general-purpose large language models in specific scientific data interpretation tasks.

Local view

Chinese tech media highlighted GPT-6 Astra's performance on the DDD Benchmark, positioning it as the strongest AI for antibody prediction following AlphaFold. Phoenix Tech (tech.ifeng.com) reported that the benchmark reveals the rigor of new testing standards and demonstrates how general large models are taking over "hardcore scientific boundaries" from small-molecule synthesis to large-molecule antibody prediction. This reflects a broader narrative in Chinese tech circles about the rapid convergence of general-purpose LLMs with specialized scientific tasks.

Context & numbers

• Mathematics: The Navier-Stokes problem carries a $1 million prize from the Clay Mathematics Institute.
• Weather Accuracy: The updated WeatherNext 3 model achieves 3-day forecast skill equivalent to previous 2-day benchmarks, effectively gaining a day of lead time for cyclone warnings.
• Materials Science: A recent Nature Communications paper discusses bridging the divide between weather and climate modeling using AI, noting that the historical separation of these fields has slowed progress in both.

On the radar

• Peer Review of OpenAI's Proof: The mathematical community is currently scrutinizing OpenAI's claim regarding the Navier-Stokes equations. Independent verification will be critical to determine if this represents a genuine solution or a heuristic approximation.
• Weather Control Research: Researchers are exploring unconventional ideas, such as using AI to predict optimal times for artificially inducing rainfall over oceans to mitigate land-based torrential downpours in Japan.