OpenAI Says Its AI Disproved a 30-Year-Old Theoretical Computer Science Conjecture

On October 26, OpenAI announced that its model had disproved an assumption from theoretical computer science that had been considered true for nearly 30 years and served as the foundation for a number of papers. What makes this story unusual is the scientists' reaction. Two researchers who originally proposed the conjecture didn't brush off the claim. They are preparing to gather colleagues and jointly verify the counterexample the AI proposed. According to Jiqizhixin, that same day the company also claimed to have solved the Kakeya conjecture, one of the well-known problems in geometric measure theory.

A second mathematical blow in a short span

Discussion of the previous batch of results hasn't ended yet. Earlier, OpenAI presented 22 mathematical manuscripts, and the community has been examining their quality, originality, and the extent of human contribution to the final formulations. The new announcement turns the conversation in a different direction. Previously, AI mostly made progress in areas requiring exhaustive search through many options or carefully completing a known proof scheme. Now the claim is that the model found the reason an established conjecture is false.

For theoretical computer science, this genre is especially sensitive. Conjectures like this often function as unspoken assumptions: conditional results of the form "if A is true, then B" are built on them. If A falls, an entire chain of conclusions must be reconsidered, and while not all of them necessarily collapse, each will need to be checked again.

Why the authors themselves will do the verification

The most interesting part is the position of those whose conjecture was challenged. The two researchers who proposed it roughly 30 years ago are organizing the verification together with colleagues. For science, this is a standard and the healthiest possible reaction: by nature, a counterexample is easier to verify than to find. If it's correct, it can be checked by hand, rewritten in independent code, or formalized in an automated verification system.

For now, this is just a company announcement, not an accepted result. The status will change once an independent group confirms the construction step by step, and mathematicians have no reason to rush here. One recent scandal involving AI-generated proofs has already taught the community to distinguish a nice-sounding formulation from a verified statement.

What's known so far

• Announcement date: October 26, source of the report — Jiqizhixin (机器之心), published October 27.
• Subject of the claim: disproof of a theoretical computer science conjecture that was widely considered true for nearly 30 years.
• Reaction of the conjecture's authors: the two researchers who proposed it are preparing a joint verification with colleagues.
• Second result: according to the outlet, OpenAI also claimed to have solved the Kakeya conjecture. It concerns sets containing a unit line segment in every direction and asserts that such sets have maximal dimension.
• Context: OpenAI had previously presented 22 mathematical manuscripts, and discussion around them is ongoing.

How to read news like this

If you work with algorithms, cryptography, or computational complexity, it's worth noting which of your results rely on the assumption now being disputed, and waiting for the verdict of independent reviewers. For AI teams, the signal is different. Mathematics is becoming a proving ground where a model can be judged not by a benchmark, but by the reaction of experts willing to check every line. If the counterexample holds up, labs will have an argument that language models are capable of finding new knowledge, not just repeating what's already known. If it doesn't hold up, that will be another lesson: an impressive result without independent verification remains a press release.