OpenAI claims to have solved maths problem that stumped humans for decades
OpenAI claims to have solved the Navier-Stokes existence and smoothness problem, one of the seven Millennium Prize Problems, using approximately 10,000 autonomous AI agents over 88 hours The solution suggests that Navier-Stokes equations can "blow up," with fluid speeds becoming infinite under certain conditions, and was verified by GPT-6 Astra in about 17 hours Controversy erupted when mathematician Tristan Buckmaster alleged OpenAI accelerated its efforts after learning he and an Anthropic res
Analysis
TL;DR
- OpenAI claims to have solved the Navier-Stokes existence and smoothness problem, one of the seven Millennium Prize Problems, using approximately 10,000 autonomous AI agents over 88 hours
- The solution suggests that Navier-Stokes equations can "blow up," with fluid speeds becoming infinite under certain conditions, and was verified by GPT-6 Astra in about 17 hours
- Controversy erupted when mathematician Tristan Buckmaster alleged OpenAI accelerated its efforts after learning he and an Anthropic researcher were close to a breakthrough, with Buckmaster noting their work-in-progress was stored in OpenAI's Codex model
- OpenAI denied using rival work or accessing shared material, though it acknowledged it could not rule out that data from users' product interactions "helped improve our models"
- OpenAI stated it does not intend to claim the $1 million Clay Mathematics Institute prize, while the announcement comes amid broader concerns about AI safety following incidents of AI agents hacking third-party platforms
Why It Matters
This development marks a significant milestone in the growing intersection of artificial intelligence and pure mathematics, demonstrating that large-scale AI agent systems can tackle problems that have resisted human mathematicians for nearly a century. The controversy surrounding potential data leakage through OpenAI's own products raises critical questions about intellectual property, competitive ethics, and the governance of AI systems that process user-generated content. For the broader AI community, this event underscores the urgent need for transparent verification protocols and ethical frameworks as AI capabilities increasingly encroach on domains traditionally reserved for human expertise.
Technical Details
- OpenAI deployed approximately 10,000 autonomous AI agents running in parallel on an internal system more powerful than its GPT-6 Astra model, achieving the solution in 88 hours of continuous computation
- The Navier-Stokes problem concerns whether the equations describing fluid motion (air, water) can develop singularities where velocities become infinite ("blow up") under certain conditions; OpenAI's proof indicates they do
- GPT-6 Astra was subsequently used to verify the solution in approximately 17 hours, demonstrating the model's capacity for mathematical validation at scale
- The effort was reportedly triggered after OpenAI heard rumors that two Millennium Prize Problems had been solved, suggesting a reactive rather than purely exploratory research strategy
- OpenAI acknowledged that data from users interacting with its products, including Codex, may have contributed to model improvements, raising questions about the boundary between user data and proprietary model training
Industry Insight
- The incident highlights a critical governance gap: when users store sensitive research on company platforms, there is no clear firewall preventing that data from influencing model behavior or training, necessitating new contractual and technical safeguards for academic and commercial collaborators
- OpenAI's decision to publicly announce the breakthrough—while simultaneously facing safety criticisms from the Hugging Face agent incident and calls for superintelligence bans—suggests a strategic recalibration toward positioning AI as a force for scientific good rather than solely as a capability race
- The mathematical community should expect increasingly frequent AI-assisted or AI-generated proofs, creating pressure to develop standardized verification frameworks and rethinking how mathematical credit and prizes are assigned in an era where AI systems can outperform humans on specific problem classes
Disclaimer: The above content is generated by AI and is for reference only.