The Riemann hypothesis, one of the toughest puzzles in mathematics, may have gotten a solution from inside Anthropic. According to New Scientist, mathematician Levent Alpöge, who works at the company, published a statement on Friday, November 20, claiming he found a proof of the problem with help from Claude, the AI built by the company itself. The key point is that the claim still needs to be checked thoroughly by experts. For now, it is a bold claim waiting for confirmation.
In the statement, Alpöge says he believes he reached the proof by working together with Claude. He also says he feels a certain sense of loss, since the problem was a kind of guiding light in his career. Even so, he says he was so impressed by what he saw during the process that he sees it as the start of something bigger for science.
To grasp how big this is, it helps to go back in time. According to the report, the hypothesis was proposed by German mathematician Bernhard Riemann in 1859 and describes a pattern in prime numbers. That is 167 years without an accepted answer. The problem is also one of the Millennium Prize Problems from the Clay Mathematics Institute, which offers $1 million to anyone who presents a recognized solution.
This is not the first time the company has poked at this problem. New Scientist notes that in August, Anthropic had already released work on the topic, but without reaching a full proof. This announcement goes much further. And it puts AI in mathematics at the center of a debate that matters to both science and the tech market.
How the proof was presented, and why it is too early to celebrate
Alpöge did not release a version that a person can read from start to finish. According to the outlet, he thought it would be inappropriate to put out a giant, unreadable document, since an unreleased version runs past 250,000 pages. What he released was the proof code written in Lean.
Here is how it works: Lean is a language in which a computer checks each logical step of a proof, one by one. This process is called formalization. If the code compiles, meaning it runs without errors from start to finish, mathematicians consider the result correct beyond any reasonable doubt.
But the method has gaps.
New Scientist itself stresses that both the code and the math behind it still need careful human review. Kevin Buzzard, of Imperial College London, reacted with caution: for him, the first step is to see whether the code actually compiles. After that, it could take time before experts can explain the line of argument. Buzzard added that the proof will probably hold up, since Alpöge would be in a very embarrassing position if the code failed or had any flaw.
If everything checks out, the impact could go beyond the original problem. Alpöge says the proof would also help with the generalized version of the hypothesis, which extends the idea to other properties of primes. Speaking to the outlet, James McKee, of Royal Holloway, said many results that relied on this version would become theorems. He noted that most mathematicians already believed, or at least suspected, that it was true, so the proof itself is not a shock. What gets to him is something else: how fast automated proofs have advanced in recent months. For McKee, it is a moment that is both wonderful and hard for people who do research in mathematics.
Behind the scenes: the timing and the track record of the author
The timing of the announcement also drew attention. Anthropic went public on November 19, and Alpöge suggests he waited for that date to pass before releasing the proof. According to New Scientist, he says he spoke with CEO Dario Amodei beforehand and understood there was a restriction to respect. The mathematician himself admitted the timing was bad and that, if he could, he would have finished everything much earlier.
Alpöge has a recent track record with AI-assisted discoveries. In July, according to the outlet, he showed that the Jacobian conjecture is false, using a counterexample of just 216 characters. He and Tristan Buckmaster, of New York University, also announced progress on three problems seen as stepping stones toward the Navier-Stokes problem, just hours before OpenAI released, amid controversy, a full solution to it.
This back-and-forth shows that the race among AI companies has reached cutting-edge math research. And fast.
What this changes for people working with AI in Brazil
If the proof holds up, artificial intelligence in research may stop being just a support tool and start sharing credit with people for top-level discoveries. That is a reading of the landscape, not a certainty. But it reinforces the idea that AI works best as collaboration rather than pure automation.
For AI in business in Brazil, the most practical lesson lies in the caution of the mathematicians themselves. Not even a code-checked proof escapes human review. The same goes for reports, code, data analysis and marketing campaigns. Picture an agency that uses AI to build a results report: if nobody checks the numbers before sending it, a small mistake can turn into a big problem with the client. It is the same reason why artificial intelligence still needs humans in the daily work of companies.
There is also a market signal, in our reading: the big AI companies seem to be competing for credibility through scientific achievements. This kind of showcase tends to carry weight when teams pick tools to optimize processes and businesses with AI. Still, an impressive announcement is no substitute for testing the tool in your own workflow.
For now, here is where things stand: a bold claim, pending verification and reactions from mathematicians that mix excitement and caution. The next steps depend on the code compiling and on experts unpacking the argument. To follow how this plays out, our coverage keeps an eye on what happens behind the scenes in the industry, including more news about Anthropic here at Método Viral.
