OpenAI’s new batch of math proofs is getting pushback from top mathematicians. This week, the company released hundreds of AI-made solutions to hard math problems. But experts say the OpenAI math proofs do not meet the standards they asked for. Here is the story in simple words.
What Did OpenAI Release?
OpenAI shared hundreds of claimed solutions to some of the hardest problems in math. The solutions came from an internal AI model that the public cannot use yet. Reports give different counts: one says 377 results, while another says nearly 720 papers. Either way, it was a very large release.
OpenAI also said it talked with a group of top mathematicians before the release. The goal was to avoid the kind of fight that followed its earlier math claims.
Who Set the Standards?
The standards come from the Advisory Group on Mathematics and AI, a group of nine researchers hosted by Princeton University. The group published its guidelines at the end of September. In short, it asked AI labs to follow these rules:
- Make sure humans can understand the proofs.
- Use computer-checkable code (called formalization) for proofs that people cannot follow.
- Share the AI’s reasoning steps.
- Give credit to earlier papers that first shared the ideas.
- Help pay for human mathematicians who check and explain the results.
Where Did OpenAI Fall Short?
According to TechCrunch and other reports, OpenAI missed several of these points. The biggest problem is human understanding. Mathematicians say a proof only becomes useful when people can read it, learn from it, and build on it.
| What Mathematicians Asked For | What Reports Say Happened |
|---|---|
| Proofs that humans can understand | Critics say the release fell short here |
| Computer-checkable (formal) proofs where people can’t follow the work | About 42% of the released proofs had not been formalized |
| Share the AI’s reasoning steps | One report says this was shared for only 10 of 719 papers |
| Link plain-language proofs to formal code with machine-readable data | Reportedly not done |
| Help fund human mathematicians | OpenAI has not yet responded, according to reports |
A Problem With the Computer Code
A new paper also found errors in how some plain-language explanations were turned into formal code. This was seen in work linked to the Navier-Stokes equations, a famous set of math problems about how fluids move. These gaps do not prove the solutions are wrong. However, they raise a question: can we trust AI models to check their own work without human help?
What Experts Are Saying
Melanie Wood, a Harvard math professor, made a key point. She said that when an AI produces a solution, the real scientific work only begins at that moment. Humans still need to read it, test it, and explain why it matters. Experts also want more openness, outside review, and teamwork between AI labs and mathematicians.
This Is Not the First Controversy
OpenAI’s math claims have caused debate before:
- October 2025: An OpenAI leader said GPT-5 had solved 10 unsolved Erdős problems. Mathematician Thomas Bloom replied that this was misleading. The AI had only found older papers that already contained the answers.
- September 2026: OpenAI announced a solution to a version of the Navier-Stokes problem. The proof was 166 pages long, and mathematicians said it was very hard to read. Still, few of them believe it is wrong.
Does This Mean the Proofs Are Wrong?
No. So far, the criticism is about how OpenAI shared its results, not about whether the math is correct. The advisory group said that mathematicians will decide whether OpenAI really followed its voluntary guidelines. Checking hundreds of proofs will likely take weeks or months.
Why This Matters
AI is getting better at math very quickly. Because of this, the field needs clear rules for how to share AI results. Without human understanding and proper checking, a proof may be correct but still teach us very little. This story shows that the math community wants AI labs to work with humans, not around them.
Frequently Asked Questions
What did OpenAI release this week?
Hundreds of claimed solutions to hard math problems, made by an internal AI model that is not yet public.
Who criticized the release?
Mathematicians, including members of the Advisory Group on Mathematics and AI, which is hosted by Princeton University.
Are the OpenAI math proofs wrong?
There is no proof that they are wrong. The main criticism is about human understanding, formal checking, and transparency.
What does “formalized” mean?
It means the proof is written as computer code that a program can check step by step.
Has OpenAI responded to the funding request?
According to reports, OpenAI has not yet responded to the call to help fund human mathematicians.
OpenAI’s math proofs show how fast AI is moving. But mathematicians are clear that speed is not enough. They want proofs that people can understand, check, and use.
