OpenAI’s latest AI systems are pushing deeper into a field once considered one of the strongest tests of human reasoning: advanced mathematics.
The company has reported a series of results on previously open mathematical problems during 2026. In August, OpenAI published ten advances spanning areas including geometry, coding theory, group theory, cryptography and theoretical computer science. The company said the results came from an internal version of its Astra model and that the arguments were subsequently formalized into Lean certificates.
The developments became even more significant in September, when OpenAI announced that an internal AI system had produced a proposed solution to the Navier–Stokes existence and smoothness problem, one of the seven Millennium Prize Problems. OpenAI said the system showed that the equations governing fluid motion can develop a singularity in finite time and released both a mathematical write-up and a formalized Lean proof.
The claim has attracted considerable attention because the Navier–Stokes problem has remained unresolved for roughly 90 years. OpenAI says its internal system was substantially more capable than GPT-6 Astra and that it used large-scale reinforcement learning on top of a pretrained model.
More Than 100 Problems?
The viral claim that OpenAI has already “solved more than 100 open problems” needs some qualification.
OpenAI has reported multiple solutions and substantial advances, but there is no single authoritative tally establishing that more than 100 previously open mathematical problems have been independently solved and accepted by mathematicians.
One prominent collection, Ben Green’s list of 100 open problems, has seen several AI-attributed developments in 2026. Independent tracking of the list records an AI-assisted negative resolution of one problem and OpenAI’s construction addressing the sofic half of another, while noting that the broader problem statuses have not necessarily been updated to “solved.”
That distinction matters because producing a convincing AI-generated argument is different from having the mathematical community independently verify and accept the result.
Why Mathematicians Are Raising Questions
The rapid progress has generated excitement as well as concerns about how AI-generated mathematics should be evaluated.
One major issue is verification. AI systems can generate extremely long and technically complicated arguments, making independent checking increasingly important. OpenAI has attempted to address this by formalizing several of its results in Lean, a proof-assistant system designed to allow mathematical arguments to be mechanically checked.
Another issue is attribution.
The Navier–Stokes announcement triggered a dispute involving NYU mathematician Tristan Buckmaster and Anthropic mathematician Levent Alpöge, who were working on related research. Buckmaster raised concerns about whether work performed using OpenAI’s tools could have influenced the company’s system. OpenAI subsequently said an internal investigation found that Buckmaster’s Codex prompts could not have influenced the system used for its result.
The controversy has contributed to broader concerns within the mathematical community about how researchers can safely use AI tools without compromising intellectual attribution or unpublished research.
OpenAI Says It Wants Greater Responsibility
OpenAI itself has acknowledged that increasingly capable mathematical systems raise questions that a technology company cannot answer by itself.
In its August mathematics publication, the company explicitly discussed its responsibility toward the mathematical community and said attribution should accurately reflect how a result was produced.
Reports also show that mathematicians and AI researchers have already been meeting to discuss the implications of increasingly capable mathematical AI. A Washington Post report from August described a gathering of prominent mathematicians at OpenAI where participants debated what the future of human mathematical research could look like as AI capabilities improve.
A New Era for Mathematical Research
The significance of these developments goes beyond individual mathematical problems.
AI systems are increasingly being used to search through possible approaches, generate conjectures, write proofs, produce counterexamples and formalize mathematical arguments. OpenAI’s May result on the Erdős unit-distance problem was particularly notable because the company described it as the first time a prominent open problem central to a mathematical field had been solved autonomously by AI.
The emerging model could therefore be less about AI simply replacing mathematicians and more about changing how mathematical research is conducted.
Humans may increasingly define questions, evaluate significance, provide mathematical context and independently verify results, while AI systems handle increasingly large portions of exploration and proof construction.
But as these systems become capable of producing research at unprecedented speed, mathematics faces a new challenge: establishing standards for verification, attribution and collaboration before AI-generated discoveries become routine.
The coming years could determine not only how many mathematical problems AI can solve, but also how the mathematical community decides what counts as a trustworthy discovery.
