On September 8, 2026, OpenAI announced that an internal artificial intelligence system situated well beyond GPT-6 Astra had solved the Navier-Stokes existence and smoothness problem. The resolution addresses one of the seven historic Millennium Prize Problems in mathematics. Across a 166-page proof formalized in the Lean verification language, the system demonstrated that singularities can develop in three-dimensional incompressible fluids within finite time. The complete mathematical proof was promptly released as open source for independent academic scrutiny.
The computational footprint required to achieve the result underlines the sheer scale of modern automated reasoning. OpenAI deployed 10,000 parallel agents that worked across approximately 88 hours of continuous compute time. Over the course of the run, the distributed network generated around 130 billion tokens to navigate and close the logical steps. This deployment highlights how brute-force agent coordination can be combined with formal verification systems to tackle previously insurmountable mathematical hurdles.
The announcement was immediately followed by sharp controversy within the mathematical community. Between September 9 and September 12, 2026, Tristan Buckmaster of New York University and Levent Alpöge of Anthropic went public with serious allegations. Both researchers had been collaborating on the exact same differential equations for an extended period. They openly accused OpenAI of steamrolling their ongoing work by leveraging massive computing power following internal leaks.
According to Buckmaster and Alpöge, their work had been entered into OpenAI Codex sessions over several months as they formulated their mathematical intermediate results. The researchers argued that telemetry or prompt exposure from these proprietary sessions gave OpenAI insight into their novel mathematical avenues. Armed with those strategic clues, the AI lab was able to deploy its vast hardware clusters and finish the proof before the original authors could publish. This dynamic has sparked intense debate over whether user inputs in coding assistants are adequately protected from competitive exploitation.
The escalating dispute drew widespread attention, prompting Fields Medalist Terence Tao to issue a public warning. Tao expressed profound concern that researchers might stop publishing working drafts or testing concepts in digital spaces out of fear of aggressive algorithmic preemption. He stressed that such defensive secrecy threatens to erode centuries of open scientific collaboration and scholarly transparency. If researchers can no longer share unfinished hypotheses without risking corporate data scraping, the fundamental mechanics of academic discovery could suffer.
The controversy marks a defining moment for the relationship between frontier AI laboratories and academic researchers. While OpenAI proved that multi-agent systems can achieve historic breakthroughs in formal mathematics, it also revealed the ethical vulnerabilities of cloud-based scientific workflows. Academic institutions are now reassessing whether working through proprietary developer interfaces poses an unacceptable threat to intellectual priority. As automated research accelerates, the boundaries of scientific attribution will require urgent, enforceable standards.