Fortune reported on July 21 that an AI system contributed to resolving the Jacobian conjecture, a mathematical problem that has remained open for roughly a century, with mathematician Levent Alpoge quoted describing the pace of change as 'very rapid and very unsettling.' That framing -- unsettling rather than purely celebratory -- is the more interesting part of the story than the mathematical result itself, and it reflects a real shift in how working mathematicians are processing AI's rate of progress on problems that have resisted human effort for generations.
This isn't an isolated data point. 2026 has produced a steady stream of AI contributions to genuinely unsolved problems across pure mathematics, physics and biology, moving well beyond AI's earlier reputation for speeding up known computations or assisting with routine proof-checking. Labs including Anthropic, Google DeepMind and OpenAI have each publicized mathematical and scientific milestones this year as proof points of frontier capability, using hard, verifiable problems as a way to demonstrate progress that's harder to dispute than benchmark scores on standard test sets.
The reaction from mathematicians quoted in Fortune's piece captures something broader happening across expert communities confronting AI capability in their own specialized domains -- a mix of genuine appreciation for a useful tool and real anxiety about what it means when systems can meaningfully contribute to the hardest open problems in a field that has always considered itself a purely human intellectual pursuit, resistant to automation in a way routine cognitive labor was not.
For AI labs, milestones like this function as recruiting and credibility signals aimed specifically at the scientific and mathematical communities whose buy-in matters for AI-for-science partnerships -- and they land at a moment when science-and-compute investing (Dimension Capital's $800 million fund, Colossal Biosciences' valuation talks) is already a hot venture category.
For founders and investors in AI-for-science, this is a reminder that the capability curve in specialized reasoning domains keeps moving faster than most experts expect, which is both the bull case for AI-native research tools and the source of the unease mathematicians like Alpoge are voicing publicly.