Alessandro Maria Masullo devoted 12 years to working on the Generalized Riemann Hypothesis. He had a manuscript containing his proof, which was currently being reviewed by mathematician Brian Conrad. Their original plan was to spend another year refining the manuscript and seeking feedback from more experts before publishing it.

This plan was invalidated on November 20th.

On that day, the mathematicians at Anthropic announced that they had used Claude to prove the Riemann Hypothesis, and claimed that the same methodology could also be applied to the generalized Riemann Hypothesis. They accompanied their announcement with a document consisting of hundreds of thousands of pages as well as formalized code. Upon hearing this news, Masullo’s team had no choice but to prematurely make the unfinished manuscript public. In their official statement, they emphasized one point in particular: this proof did not make use of any AI tools; it was entirely completed by human beings.

Why is there such a rush to publish? In the mathematical community, priority is determined by the time of publication. If one waits for a year before publishing, even if the two approaches to proving the theorem are completely different or even if the manuscripts have actually existed for a long time, they will still be listed as “the second” contribution in the official records. By publishing earlier, at least one timestamp is created, indicating that this approach had already been largely developed before the AI-related announcement was made.

This is not the first time something similar has happened this year. In September, when OpenAI announced that it had solved the Navier-Stokes problem, there was considerable controversy. At that time, several other mathematicians were also working on this problem and had already made significant progress.

When these two events are considered together, a question that has never been seriously discussed before becomes apparent. A mathematician spent over a decade working on a major conjecture, following the traditional process of submitting the manuscript for review, making revisions, and then finally publishing it. This approach assumed that competitors were also human beings, and that their progress would be roughly comparable. However, now the competition involves systems that are capable of producing hundreds of thousands of pages in just one and a half months; as a result, taking too much time to refine such an approach has become risky.

Masullo’s manuscript is currently in a state where it has not been thoroughly reviewed; whether this is indeed the case will have to wait for peer evaluations. If both of these claims prove to be true, it would constitute a rather unusual situation: in this case, there would be a conjecture that one person is able to read but has not yet completed the review of, while another version of the same conjecture has been examined by a machine but remains incomprehensible to anyone.