OpenAI Claims AI Breakthrough on Million-Dollar Math Challenge Sparks Controversy

Key Points

• An experimental system involving approximately 10,000 AI agents reportedly resolved a portion of the Navier-Stokes mathematical challenge within 88 hours
• During the computational process, these agents communicated through 2.7 million messages and produced approximately 130 billion tokens
• The underlying AI model surpasses the capabilities of any system OpenAI has made available to the public
• Tristan Buckmaster, a mathematics professor at NYU, contends that OpenAI might have accessed details from his ongoing, unpublished investigation
• The company has stated it will not pursue the $1 million award, and the proposed solution awaits independent validation

OpenAI has announced that a large-scale experimental deployment of approximately 10,000 AI agents successfully addressed a component of the Navier-Stokes existence and smoothness challenge within an 88-hour timeframe. This mathematical puzzle represents one of the seven prestigious Millennium Prize Problems, each accompanied by a $1 million bounty from the Clay Mathematics Institute.

The Navier-Stokes equations govern the behavior of fluid dynamics, including liquids and gases. The unresolved mathematical question centers on whether initially smooth fluid motion can eventually deteriorate, resulting in velocities that escalate infinitely. OpenAI’s AI system asserts this phenomenon is possible.

The proposed mathematical proof outlines a fluid vortex structure that elongates and intensifies until its velocity becomes unbounded within a limited timeframe, despite the system maintaining finite overall energy.

Throughout the problem-solving endeavor, the AI agents conducted approximately 2.7 million message exchanges and produced roughly 130 billion computational tokens. Calculating based on OpenAI’s standard pricing structure, this magnitude of computational resources would amount to approximately $10 million in costs.

The technology powering this achievement is a proprietary system that OpenAI describes as superior to GPT-6 Astra, currently its most sophisticated publicly accessible model. Following the agents’ completion of their collaborative work, Astra dedicated an additional 17 hours to verification and formalization of the mathematical proof.

According to OpenAI’s timeline, training for this advanced model commenced in late August. On September 1, the organization received informal reports suggesting that two Millennium Prize challenges had been resolved by outside parties. This prompted OpenAI to direct its AI agent network toward the outstanding problems.

By September 5, approximately 88 hours after initiation, OpenAI had generated a candidate solution.

Academic Researcher Questions Potential Use of Unpublished Findings

The disclosure prompted an immediate reaction from Tristan Buckmaster, a mathematics faculty member at New York University. He revealed that he and Levent Alpöge, a mathematician employed by Anthropic, had been investigating a substantially similar problem for roughly twelve months.

Buckmaster disclosed that both researchers had utilized OpenAI’s Codex platform during their investigative work. On September 3, he became aware that details regarding their research advancement had been communicated to OpenAI.

According to Buckmaster, OpenAI commenced its work on the Navier-Stokes challenge only after receiving this information. He emphasized that virtually no other mathematicians were exploring the identical methodology that he and Alpöge had adopted.

OpenAI has refuted any direct access to the work produced by Buckmaster and Alpöge. The organization stated that no user information was extracted and that it cannot verify any linkage. Nevertheless, the company acknowledged it cannot completely eliminate the possibility that data generated through their utilization of OpenAI’s services may have contributed to model enhancement indirectly.

OpenAI extended recognition to both researchers for their contributions, characterizing the work as simultaneous rather than preceding. The company additionally noted that the two mathematical proofs are distinct and actually establish marginally different conclusions.

The proposed solution remains unverified by independent authorities. The Clay Mathematics Institute mandates that candidate solutions undergo extensive peer review spanning multiple years before prize consideration.

OpenAI has indicated it will not pursue the $1 million prize, positioning the achievement instead as evidence of the accelerating advancement rate of its research infrastructure.