The Post-Quantum Security Model Built for an Age of AI Agents

A Ground-Breaking Framework that Could Permanently Change How We Work with Cryptography in the Quantum Age

On 5 September 2026, a proof appeared.

It had been ninety years in the making.

It answered one of the seven Millennium Prize Problems — the Navier–Stokes existence and smoothness problem, a question about whether the equations governing fluid flow can develop a singularity in finite time.

Human mathematicians had not solved it.

An AI system did.

Not one model working alone.

Roughly ten thousand agents running concurrently, exchanging nearly 2.7 million messages, burning through compute at a cost OpenAI’s own Chief Research Officer described as in the millions of dollars.

Eighty-eight hours.

Then GPT-6 Astra formalised the proof in Lean — a system that automatically verifies mathematical proofs step by step — in seventeen additional hours.

OpenAI published the full result on 8 September 2026, framing it carefully as evidence still requiring mathematician review rather than a claim to the prize.

Two human researchers, Tristan Buckmaster and Levent Alpöge, had concurrent work on the related forced Euler equations using an internal Anthropic model, and OpenAI recognised their priority.

Science reported the full story, including the priority dispute that followed.