AI and open math problems in 2026: what was claimed, what holds up
OpenAI claimed a Navier–Stokes proof and posted about 720 AI-written manuscripts in 2026. Mathematicians accept some results and dispute others.
AI mathNavier-Stokes AIErdős problems AIOpenAI mathTerence Tao AIAI research jobs
Math is where AI capability is easiest to check, which is why 2026's results there are worth reading carefully. This is what was claimed, what mathematicians say about it, and what it signals for research and technical work. Claims are dated as of Nov 17, 2026; the hiring data runs to Oct 30, 2026.
What was claimed
| Date | Claim | Status |
|---|---|---|
| May 20, 2026 | OpenAI reported a model disproved Erdős's 1946 planar unit distance conjecture | Checked by external mathematicians, per OpenAI |
| Jul 2026 | AI systems reported 42 of 42 at the International Mathematical Olympiad | Company statements reported by AFP |
| Sep 8, 2026 | OpenAI claimed a proof of blow-up for 3D Navier–Stokes with a smooth force | Lean proof compiles; human review ongoing; the unforced case is still open |
| Oct 6, 2026 | OpenAI published about 720 AI-written manuscripts on open problems | About 42% formalized in Lean; several disputed |
Source: Jobs Demand Index, data to Oct 30, 2026.
What mathematicians say
The Clay Mathematics Institute said the Navier–Stokes problem has "apparently been settled" and that its review is deliberately unhurried. Mathematicians quoted by NPR said the 166-page proof teaches little so far, and Tristan Buckmaster of NYU disputes how credit was handled. On the October manuscripts, Frank Calegari and Buckmaster said many are poorly written and some are wrong. Terence Tao warned that good open problems are being used up in a non-renewable fashion.
So the picture is mixed in a specific way. Machines now produce correct results on hard, well-posed problems quickly. Explaining them, judging which ones matter and taking responsibility for them still falls to people.
What it signals for work
Dwarkesh Patel, who earlier in 2026 put superintelligence most of a decade away, now writes that AI matching the best AI researchers may be a couple of years off. His essay's title is the point: the avalanche that hit math is coming for other fields.
| Occupation | US hiring demand | Sector postings | BLS growth 2025–35 | Median pay | Live roles in our sample |
|---|---|---|---|---|---|
| Mathematicians | Steady | +3.9% | +0.6% | $126,710 | Too few to report |
| Statisticians | Rising | +3.9% | +11.0% | $105,650 | Too few to report |
| Physicists | Rising | +20.9% | +7.2% | $172,250 | Too few to report |
| Electronics engineers, except computer | Rising | +34.1% | +3.7% | $130,220 | 175 at 36 companies |
| Computer and information research scientists | Rising | +23.8% | +21.8% | $140,300 | 406 at 101 companies |
| Mathematical science teachers, postsecondary | Cooling | −8.8% | +1.7% | $79,940 | Too few to report |
Source: Jobs Demand Index, data to Oct 30, 2026. Sector postings are measured for the whole Indeed sector an occupation sits in, year on year. Median pay: BLS OEWS 2025. Live roles: public company job boards, Nov 3, 2026 pull.
Four of these six are rated Rising in the US, one Steady and one Cooling. The hiring data does not show the shift yet. That is the lag. Our reading: in research work, the leverage moves from producing a result to choosing the question, checking the answer and standing behind it.
What to do with this
Check your own occupation's demand and observed AI use in the occupations search, and follow dated milestones on the AI Frontier page.
Limits: these are claims by companies and commentators, reported with their dates. The hiring figures are sector-level postings and BLS projections.
Check the occupations in this article
Each page shows US hiring demand and the evidence behind it, the Canadian sector postings trend and provincial outlooks shown separately, and a decision panel for your next move. Related: questions people ask, market charts, build your own chart.
Sources
- Clay Mathematics Institute, Navier-Stokes announcement, Sep 11, 2026 (claymath.org)
- NPR via Georgia Public Broadcasting, AI solved one of math's hardest problems. Humanity learned nothing (so far), Sep 22, 2026 (gpb.org)
- The Tufts Daily, Mathematicians still checking the Navier-Stokes proof, Sep 24, 2026 (tuftsdaily.com)
- OpenAI, math repository README, Apache 2.0, retrieved Nov 17, 2026 (github.com/openai/math)
- Let's Data Science, OpenAI posted 722 AI-written math papers, Oct 2026 (letsdatascience.com)
- Quanta Magazine, How AI tore through a mathematical community, Aug 3, 2026 (quantamagazine.org)
- Terence Tao, post on Mathstodon, Sep 8, 2026 (mathstodon.xyz)
- AFP via The Standard, AI catches up with humans to score 100pc at top maths contest, Jul 23, 2026 (standardmedia.co.ke)
- Dwarkesh Patel, The avalanche that hit math is coming for AI development, Nov 13, 2026 (dwarkesh.com)
Cite: Jobs Demand Index, "AI and open math problems in 2026: what was claimed, what holds up", Nov 17, 2026 (data to Oct 30, 2026). https://jobsdemandindex.com/insights/ai-open-math-problems-2026-claims. How ratings are built. Plain-text version.
Related insights
AI and engineers: hiring is rising fastest for those who build hardware
Two in three live tech roles mention AI. The steepest growth in US job postings is in mechanical, industrial and electrical engineering.
Semiconductor and photonics engineers: what 1,763 live roles show
Electrical engineering postings are up about 34% in both the US and Canada, and 1,763 live roles in our sample are semiconductor work.
Data engineering jobs: postings 37% below 2020, outlook still strong
US data and analytics postings grew 3.9% in a year, close to the all-jobs pace, and remain far under their 2022 peak. The long-run outlook points up.