Cardano founder weighs OpenAI’s math breakthrough

Cardano founder Charles Hoskinson said on Sept. 9 that artificial intelligence had advanced further in formal mathematics than he expected.

Hoskinson said artificial intelligence's mathematical progress considerably exceeded his earlier expectations publicly.

OpenAI says roughly 10,000 agents produced a Navier--Stokes solution within 88 hours of work total.

Clay still classifies Navier--Stokes as unsolved pending publication, review and broad mathematical acceptance worldwide today.

Hoskinson raised confidentiality concerns for researchers entering unpublished work into centralized cloud AI systems online.

OpenAI denied accessing private work but could not exclude de-identified usage data influencing model improvements.

His comments followed OpenAI's claim that an internal system produced a solution to the Navier--Stokes Millennium Prize Problem.

During a broadcast, Hoskinson called the reported capabilities "pretty remarkable." However, he also addressed unresolved questions about the work's provenance and the privacy of research submitted to cloud-based AI services.

Hoskinson said he originally expected formal systems to help larger teams of mathematicians collaborate and verify human-written proofs. He did not expect large language models to generate complete proofs themselves so soon.

"We never anticipated the extent to which AI would come in," Hoskinson said. He added that the idea of AI fully writing a proof had previously appeared "pretty far out."

Hoskinson has a direct connection to formal mathematics research. In 2021, he donated $20 million to Carnegie Mellon University to establish the Hoskinson Center for Formal Mathematics, according to the university's announcement.

His latest comments also fit Cardano's broader experimentation with artificial intelligence. As crypto.news previously reported, Hoskinson has defended Cardano AI agent experiments involving communications, community activity and the privacy-focused Midnight ecosystem.

OpenAI published its research on Sept. 8. The company said an internal model coordinated roughly 10,000 agents and produced a proposed solution after 88 hours. GPT-6 Astra then spent another 17 hours formalizing and checking the argument in Lean.

The proof attempts to establish that an initially smooth, stationary fluid can develop a singularity in finite time when subjected to a smooth external force. OpenAI said this satisfies statements C and D in the official Millennium Prize formulation.

The company also released an analytical paper and Lean code. A Lean formalization provides machine-checkable verification that the encoded steps follow from the stated assumptions. It does not independently establish that every definition and assumption accurately represents the intended mathematical problem.

OpenAI said it does not plan to seek the associated $1 million prize. The company nevertheless described its work as a resolution of the problem.

The Clay Mathematics Institute still labels the Navier--Stokes problem "unsolved." Its website had not recognized OpenAI's proposed proof as an accepted solution at the time of reporting.

Clay does not accept proposed solutions through direct submissions. Under its rules, a solution must appear in a qualifying publication. At least two years must then pass, and the work must gain general acceptance from the global mathematics community.

That process means OpenAI's announcement and formal proof do not constitute immediate institutional recognition. Mathematicians must examine whether the construction satisfies the precise problem statement and whether its use of external forcing answers the question as commonly understood.

The announcement also drew scrutiny involving New York University mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge. The researchers had been working on a related Euler-equation result using a forcing approach.

Buckmaster questioned whether private work entered into OpenAI's Codex system could have contributed to the company's result. He stopped short of alleging proven misconduct, saying: "I do not know whether our data was used."

OpenAI denied accessing their specific work. However, the company said it could not completely rule out the possibility that de-identified data from their product use had helped improve its models. OpenAI maintained that its proof was developed independently and differed from the researchers' work.

Hoskinson argued that the dispute should concern researchers handling unpublished ideas. He said scholars using centralized AI services should consider whether prompts, notes and research logs remain confidential.

The next phase will involve public examination of OpenAI's paper and Lean formalization. Until specialists review the assumptions and Clay's formal conditions are met, the work remains a claimed solution rather than a recognized resolution.