Which Millennium Prize Problem will AI solve next?

The current setup is important: this market is not asking whether the Birch and Swinnerton-Dyer Conjecture will be solved eventually, but whether it will be the first Millennium Prize Problem that an eligible AI lab publicly announces as solved by the end of 2027. That makes the event much more demanding than a simple math-breakthrough forecast. The news context does not show any reliable evidence of a BSD proof, any near-term announcement from an eligible lab, or even a standout claim that BSD is the front-runner among the five qualifying problems. In that sense, the present information supports caution rather than enthusiasm.

The strongest argument for Yes is that AI has recently shown real mathematical capability, including helpful results on open problems and increasingly credible proof-generation workflows. If those trends compound quickly, it is conceivable that an AI lab could produce a landmark proof in a deep number-theory area like BSD, especially if the method builds on existing partial theory and formal verification tools. However, BSD is an especially demanding target: it sits at the intersection of elliptic curves, L-functions, and deep arithmetic structure, and progress has historically been difficult even for top human mathematicians. The fact that recent coverage discusses AI progress broadly, but not BSD specifically, suggests that the path from general math competence to a BSD solution remains long.

Another key consideration is competition across the five eligible problems. Even if AI produces a major breakthrough before 2027, the first solved problem may not be BSD. Problems like P versus NP or the Riemann Hypothesis may receive more attention and more concentrated computational or theoretical effort, and a breakthrough there would beat BSD in this market even if BSD were eventually solved later. The market’s current pricing around 21% implies some belief in a nontrivial chance of an AI-led Millennium breakthrough, but the absence of any credible BSD-specific lead means that most of that probability should be reserved for the broader category of a first AI solution elsewhere, not for BSD itself.

There is also a procedural hurdle baked into the rules: an eligible AI lab must publicly announce the solution in a way that clearly credits the lab’s own work and does not rely on outside researchers’ credit. Even if an AI-assisted proof exists privately or is published by a mixed team, it may not qualify here unless the announcement satisfies the market’s strict attribution and timing requirements. That reduces the practical odds of BSD winning this race, because the event rewards not just mathematical success but a very specific kind of public, qualifying disclosure.