AI workloads, data platforms, and infrastructure notes, written from the engineering edge between benchmarks and production.

RSS feed
/

AI May Have Just Solved Math’s Greatest Mystery. Here’s Why It Matters to Everyone.

Anthropic and Claude may have proved the Riemann hypothesis—what the 167-year puzzle is, why primes and encryption matter, and what to watch as experts verify the claim.

I

Itzik — VP Mission Alignment, VAST Data

·

·

11 min read


“If I were to awaken after having slept for a thousand years, my first question would be: Has the Riemann hypothesis been proven?”

— David Hilbert, the German mathematician who in 1900 put the Riemann hypothesis on his famous list of the 23 greatest unsolved problems in mathematics (source)

Riemann hypothesis · Claude AI proof · Anthropic · why it matters beyond math

On Friday, November 20, 2026, a mathematician at Anthropic announced that he and Claude, the company’s AI, may have proved the Riemann hypothesis using Claude AI.

If you’ve never heard of it, you’re not alone. But inside mathematics, this is the big one. It’s a puzzle the brightest minds on Earth have failed to crack for 167 years. It carries a $1 million prize. One leading expert called the claim potentially the biggest thing ever to happen in mathematics.

The proof still needs to be checked, and there may be bumps along the way. But even the fact that an AI could make a serious attempt tells us something about where the world is heading. Let me explain what the Riemann hypothesis is, what happened, and why it matters, with no math degree required.

The Riemann hypothesis, without the math

Start with prime numbers: 2, 3, 5, 7, 11, 13 and so on. They can only be divided by 1 and themselves. Every other whole number is built by multiplying primes together, the way molecules are built from atoms. They also protect your bank login: much of the encryption on the internet leans on how hard primes are to untangle.

The strange thing about primes is that they look random. Sometimes two sit side by side, like 11 and 13. Sometimes there is a long gap with none at all. For thousands of years nobody could say exactly how they are spread out.

In 1859, a German mathematician named Bernhard Riemann found a clever way in. He built a mathematical “machine” that turns the list of primes into a set of special points, called zeros. Think of the zeros as the musical notes the primes are secretly playing. Riemann guessed that every important zero sits on one single straight line.

Simplified diagram of the Riemann hypothesis linking prime numbers to zeta function zeros

Simplified illustration, not to scale

If he is right, the primes are as orderly as they could possibly be. Their randomness is the well-behaved kind, like a fair coin, never wildly lopsided. Computers have checked trillions of zeros, and every one sits on the line. But checking is not proving. There are infinitely many zeros, and one stray zero anywhere would break the rule. That is why a proof has been worth a $1 million prize from the Clay Mathematics Institute since 2000, and why it has been called the most important open problem in mathematics.

A 167-year manhunt

Illustration of how prime numbers underpin internet encryption and why the Riemann hypothesis matters

Key moments in the hunt for a proof, not to scale

Riemann tossed out his guess almost casually, in a short paper of about eight pages. He wrote that he had tried to prove it, gave up for the moment, and moved on. He died seven years later, at 39. The guess outlived him by more than a century and a half.

In 1900, David Hilbert, the most influential mathematician of his day, gave a famous speech listing 23 problems for the new century. The Riemann hypothesis was number eight. Most of the other problems on that list have since been solved. This one hasn’t.

In 1914, the British mathematician G.H. Hardy showed that infinitely many zeros sit on Riemann’s line. That sounds like the end of the story, but it isn’t. “Infinitely many” is not the same as “all”. There could still be infinitely many more somewhere else.

Then came computers. Alan Turing, the father of computer science, spent time in the early 1950s using one of the first electronic computers to hunt for a zero off the line. He didn’t find one. Nobody has since, even after checking trillions.

In 2000, the Clay Mathematics Institute named seven “Millennium Prize Problems”, each worth $1 million. Only one has been solved so far: the Poincaré conjecture, cracked by the Russian mathematician Grigori Perelman, who famously turned down the money.

Along the way, many people have announced proofs of the Riemann hypothesis. Some were amateurs. Some were giants of the field: in 2018, the celebrated mathematician Sir Michael Atiyah presented a proof at a conference, and other experts quickly found it didn’t hold up. That history is exactly why mathematicians are reacting to this week’s news with excitement and a healthy dose of caution.

What actually happened on November 20

Levent Alpöge, a number theorist at Anthropic, posted a short four-page note on his personal website. In it he wrote that he believes he has, working with Claude, discovered a proof of the Riemann hypothesis. He says the same method also proves a broader version, the “generalized” Riemann hypothesis.

What he shared is unusual. Instead of a normal paper, he released code written in Lean, a computer language that lets a machine check a proof step by step. Science magazine reports it runs to about 17 million lines. The human-readable explanation, which Claude also wrote, is said to be more than 250,000 pages long and has not been released.

That matters, so let me be clear about what is and isn’t known yet:

•Not yet confirmed. No independent expert has signed off. Mathematicians still need to check that the code proves exactly the famous statement, not something that only looks like it.
•Hard to read. Even if it is right, no human can read a quarter of a million pages quickly. Understanding why it works may take years.
•Not out of nowhere. It reportedly builds on an AI result OpenAI announced in October, which proved a weaker “quasi” version of the same hypothesis.

The reaction from experts mixes wonder with caution. Brian Conrey, one of the world’s leading Riemann specialists, called it a monumental moment in human history. Others, quoted in Nature, stress that it is historic only if it fully checks out. That “if” is doing real work, and an honest blog post should say so.

How does an AI actually “do” math?

It’s natural to picture a robot scribbling equations on a chalkboard. The reality is closer to a very fast, very patient research team with a strict referee standing behind it.

A mathematical proof is a chain of small, logical steps. Each step must follow from the one before, like dominoes lined up across a room. If one domino is missing, the whole chain fails. For hard problems, the chain can run to thousands of steps, and humans make mistakes along the way. Famous proofs have been announced, celebrated, and then quietly withdrawn when someone found a missing domino.

Modern AI systems are good at two things that help here. First, they can try a huge number of ideas quickly: this approach, that trick, a technique borrowed from a paper written decades ago. Most attempts fail, and that’s fine, because the AI doesn’t get tired or discouraged. Second, they can write their reasoning in a form a computer can check.

That second part is where Lean comes in. Think of Lean as a spell-checker for logic, except far stricter. It doesn’t care how confident the author sounds or how famous they are. It checks every single step, and if one doesn’t follow, it refuses to accept the proof. That’s why the 17 million lines of code are such a big deal: they are the part that doesn’t rely on trust.

The catch is that a checker can only confirm that the steps are valid. It can’t tell you whether the final statement is exactly the famous one people care about. Experts still have to read the very first and very last lines and say: yes, that is the Riemann hypothesis. That’s the step happening now.

Wait, does this break internet security?

This is the first question many people ask, because primes protect so much of our online lives. The short answer: no.

Much of today’s encryption relies on a simple fact. Multiplying two huge prime numbers together is easy, but taking the result and working out which two primes were used is incredibly hard. Even the fastest computers would need an impractically long time.

The Riemann hypothesis is about how primes are spread out on average, not about a shortcut for finding the two hidden primes inside a specific number. Proving it tells us the primes behave the way mathematicians have long expected. It doesn’t hand anyone a master key.

In fact, many computer scientists already assumed the hypothesis was true when designing and analyzing algorithms. A proof mostly confirms that the ground they were standing on was solid all along. Your bank account is as safe on Saturday as it was on Thursday.

Why should you care if you don’t care about math?

Honestly, a proof about primes won’t change your grocery bill next week. The real news isn’t the answer. It’s what produced it.

For years, the common criticism of AI was that it was a fancy autocomplete. It could write a pleasant email, but it couldn’t truly reason: hold a long, careful chain of logic together without making things up. A proof of the Riemann hypothesis is the opposite of making things up. Every single step must be correct, and here a computer checks every step.

If an AI can do that on the hardest problem humans have, the same skill points at problems that touch everyday life:

•Medicine: reasoning through millions of possible molecules to find a drug candidate faster.
•Engineering: designing chips, batteries and materials, then proving the design works before anything is built.
•Software: writing code that is mathematically checked to be free of whole classes of bugs, in planes, cars and banks.
•Science: testing ideas in physics or biology that would take a human team decades.

I work in AI infrastructure, the storage and data systems these models run on. From where I sit, the pattern this year is clear. Results that experts said were a decade away keep arriving in months. Riemann may just be the loudest one yet.

The human side: wonder, and a little loss

Not everyone is celebrating, and it’s worth understanding why.

For many mathematicians, the Riemann hypothesis is more than a puzzle. It’s a lifetime companion. People have spent entire careers circling it, proving small pieces, hoping to see it solved before they die. Imagine training for the summit of Everest your whole life, and then watching a drone fly to the top. You’d be amazed. You might also feel something you can’t quite name.

The press coverage this week has captured that mix well: wonder that it may finally be done, and grief that it wasn’t done by a person, in a way a person can fully follow. A 250,000-page explanation is not something you read over coffee. If the proof holds, humanity may know that the hypothesis is true long before it truly understands why.

I think that feeling will spread well beyond mathematics. Lawyers, doctors, engineers and writers will all have their own “drone at the summit” moment in the coming years. Some of it will feel like losing something.

But here’s the hopeful part. Understanding the proof will itself become a huge human project, like explorers mapping a new continent someone else just spotted. Choosing which problems are worth solving, judging whether an answer matters, and explaining it to the rest of us will remain deeply human jobs. The tools changed. Curiosity didn’t.

What to watch over the next few months

If you want to follow this story without a math degree, here are the signals that matter:

•Independent checks of the code. Expect experts around the world to run the Lean files on their own machines and confirm the final statement is really the Riemann hypothesis. If several respected teams agree, that’s a very strong sign.
•A readable version. Watch for a shorter, human-friendly explanation, whether written by people, by AI, or both. That’s when the proof turns from a result into knowledge others can build on.
•The prize committee. The Clay Institute has rules, including waiting periods and expert review. Don’t expect a quick decision, and don’t read silence as doubt.
•More “impossible” results. This year has already brought a string of AI breakthroughs in mathematics, including the weaker “quasi” version in October. If the pace continues, more long-standing problems may fall, likely in physics, chemistry and computer science too.

The simplest test is this: a year from now, are more people using AI to do things that used to be impossible, or fewer? Every trend I see from inside the industry points one way.

Bumps and all, AI is going to change the world

Here is what I take from this week.

The proof may still turn out to have a gap. Experts will need months, maybe years, to be sure. If it does fall apart, it will join a long list of famous near-misses, and the headlines will swing from “AI solves math” to “AI overhyped”. That is what the bumps look like. Every powerful technology has them: exciting claims, embarrassing mistakes, skeptics who are sometimes right and sometimes very wrong.

But zoom out. A year ago, the idea that an AI could even attempt the most famous open problem in mathematics sounded like science fiction. Now the debate is about how to check its answer. When the argument shifts from “can it?” to “is it right?”, the world has already changed.

David Hilbert said that if he woke after a thousand years, his first question would be whether the Riemann hypothesis had been proven. He didn’t need to sleep that long. Barely a century after he asked, the answer may be arriving, and it may have come from a partnership between a human and a machine. Whatever happens next with this proof, that partnership is here to stay.

Related on Lots of Data: Anthropic Scion and AI Infrastructure · AI-Enabled Cyberattacks · Grok Bot and Always-On Agents