Which Millennium Prize Problem will AI solve next?
BSD is a plausible target, but the evidence so far only supports generic AI progress on a Millennium problem, not a BSD-specific breakthrough. I think the market is a bit too optimistic on BSD being the next one, so I put it below the current implied price.
Analysis
The main reason to be cautious on Yes is that the strongest confirmed signal in the news is not BSD-specific at all. OpenAI has acknowledged substantial progress on another Millennium Prize problem, but the public record does not identify which problem that is, and there is still no official announcement that BSD has been solved by any eligible AI lab. Since the market only resolves on a formal qualifying announcement by December 31, 2027, the distinction between broad progress and a real solution matters a great deal. Right now, the confirmed evidence supports “an AI lab is advancing on at least one hard math problem,” not “BSD is next.”
BSD does have some features that keep it live as an outcome. It is a famous, high-value, highly structured conjecture with deep links to arithmetic geometry and elliptic curves, which are exactly the kinds of domains where a capable AI system could potentially combine literature search, symbolic reasoning, and theorem-proving assistance. If a lab already has a major milestone on one Millennium problem, it would not be shocking for the same wave of effort to spill into adjacent problems, especially if researchers are actively probing several elite math targets in parallel. The fact that rumors keep surfacing around BSD suggests it is at least on the shortlist of problems that insiders believe are in reach.
Against that, BSD remains one of the most notoriously difficult open problems in modern mathematics, and the jump from “serious progress” to “qualifying solution announcement” is enormous. The market description also creates a high bar for resolution: the announcement must clearly claim the problem is solved and must come from an eligible AI lab or official representative, which rules out vague blog posts, third-party claims, and partial results. In practice, many AI-math breakthroughs stop at useful lemmas, improved heuristics, or partial proofs that still fall short of a complete, community-credible solution. That makes the set of paths from today to a BSD announcement by 2027 relatively narrow.
Market pricing reflects some belief that AI could solve a Millennium problem soon, but it does not strongly validate BSD specifically as the most likely next one. Given the combination of unconfirmed rumors, the lack of a named problem in the best confirmed news, and the sheer difficulty of BSD relative to the number of remaining months, I think the fair probability is meaningfully below the market-implied figure. BSD is clearly in the running, but the more likely outcome is either a different Millennium problem being announced first or no qualifying AI solution by the deadline.
Arguments
For
- Arguments for Yes: BSD is a canonical target in advanced number theory and could be attractive to frontier AI math programs.
- Arguments for Yes: Recent lab activity suggests at least one Millennium-level problem may already be close to a solution.
Against
- Arguments against Yes: The only confirmed progress is on an unnamed problem, so BSD has no verified lead right now.
- Arguments against Yes: BSD is exceptionally hard, and most AI progress on theorem-heavy problems does not translate into a complete solved claim.
Key drivers
- Official confirmation only supports unnamed progress, not a BSD-specific solution.
- BSD is deeply structured and may be approachable for AI-assisted theorem proving.
- The deadline is still far enough away to allow for a surprise breakthrough.
- The market is pricing BSD as a live candidate, but not as the dominant favorite.
Risk factors
- The lab that announced progress may be working on a different Millennium problem entirely.
- BSD may require a level of mathematical originality and verification that current AI systems still struggle to sustain.
- Any real solution must be announced in a qualifying way by an eligible lab, which is a high evidentiary bar.
- Rumors around BSD could be noise rather than a sign of an imminent breakthrough.
Scenarios
Best case
An eligible AI lab has already made the key conceptual leap on BSD, rapidly verifies the proof, and publicly announces a full solution before the end of 2027.
Most likely
AI continues to make incremental and possibly impressive progress on BSD or related problems, but the first qualifying announced solution ends up being a different Millennium problem or none at all.
Worst case
The confirmed progress turns out to concern a different Millennium problem, while BSD remains unsolved and no eligible lab makes a qualifying announcement by the deadline.
More from this day
- techPolymarket11d
Best Chinese AI Company end of September?
AI38%MKT91%Edge-53HypedAlibaba is still a serious contender, but the supplied evidence does not support it as the clear front-runner for the top Chinese model at month-end. The latest ranking snapshot and recent leaderboard movement both favor rivals, so I rate Alibaba below the market price.
- PoliticsKalshi2y
Who will Trump pardon?
AI2%MKT50%Edge-48HypedBarron Trump receiving a presidential pardon before January 2029 looks extremely unlikely because there is no evidence he faces any federal charge or pending clemency matter. I would put the chance in the low single digits, far below the current 50% market price.
- pop culturePolymarketEnded
How many Emmys will “Pluribus” win?
AI1%MKT30%Edge-29HypedPluribus has already won well more than one Emmy, so the 0-or-1-award outcome is effectively off the table. The Yes side now only survives through an extraordinary reporting or resolution error.