Key Highlights
- OpenAI released more than 700 academic papers claiming progress or solutions on 372 problems that had baffled scholars for decades.
- The Association for Human Mathematics, over 900 scholars, said mathematicians did not ask for this work.
- AI systems already improve weather forecasts, propose millions of candidate materials and help control nuclear fusion experiments.
- Anthropic has already connected its models to laboratory equipment, giving AI direct reach into physical experiments.
A Single-Day Release
OpenAI, an American frontier AI lab, released more than 700 academic papers on October 6, 2026. The papers claim progress or outright solutions for 372 problems that had baffled scholars for decades. Mathematicians are still working through what the release means for their discipline.
The effect has reached well beyond maths. The episode has become a test case for how AI changes science as a whole: it is at once a leap in what machines can do and a source of unease about a future in which understanding of the universe sits with systems that humans cannot follow step by step.
Why Maths Came First
Maths has shown the potential of AI most vividly because its proofs are derived from axioms rather than from evidence. In principle, a proof can be checked mechanically, which makes the field unusually suited to machine contributions that can be verified without trusting the machine.
Other sciences lean on maths as a tool, so progress there has carried over. AI has improved weather forecasts. It has proposed millions of possible new materials and helped control experiments in nuclear fusion. It can also predict the shapes of almost every known protein from the amino acids that make them up, an advance that earned a Nobel prize.
Scholars Push Back
The reaction among mathematicians has been sharp. The Association for Human Mathematics, a group of more than 900 scholars, said that mathematicians did not ask for this work to be done. Its members argue that the reward in mathematics lies less in the finished proof than in the effort of producing it, because that effort is what advances human understanding.
On that view, receiving correct but incomprehensible solutions from an AI resembles being handed the answers to homework: the result exists, but the learning does not. Many mathematicians have also objected to a lab entering their discipline in this way.
Critics of that position say the complaint contains an element of special pleading. They note that no profession holds an exclusive right to the production of knowledge, and that academic hierarchies built on proving theorems are not fixed. Supporters of the machine approach also point out that science already relies on mechanisms that work without being fully understood, as with many common medicines.
Biology Is the Next Frontier
The most active area outside maths is biology. Researchers are using AI to design drought-resistant crops, to search for new drugs and to create viruses that attack harmful bacteria. Another effort aims to build computer models of how cells work.
With such models, scientists gain a route toward the ability to create entirely new organisms from scratch. In that picture, DNA becomes another form of computer code, handled like a string of tokens in an AI model. The prospect draws both excitement and alarm.
Why Labs Invest in Science
AI companies are putting resources into science for reasons that go beyond curiosity. As with human learners, solving hard problems can make models more capable. Many researchers think a model that improves at scientific problem-solving improves at other tasks too.
If that holds, the largest labs gain an advantage that smaller competitors find hard to match, because they control models that can work on improving themselves. These systems reach the physical world through humanoid robots or by being connected to laboratory equipment, a step Anthropic has already taken.
The arrangement raises a practical question for research funding. People have to supply the resources, including laboratories and computing power, for machines to keep making progress. Where no one understands what the systems are doing, the question of who directs and monitors that spending becomes pressing.
Checking and digesting a computer’s results can look like unglamorous work. Much human-produced science already contains errors because scholars are rarely rewarded for policing the work of others. The October 6 release has turned attention to who verifies machine output, how that effort is credited, and whether labs submit their methods and code to peer review.






Please wait processing your request...