Skip to content
LIVE
The Executives BriefThe Executives Brief

Anthropic mathematician claims Claude cracked the Riemann hypothesis. Math is not ready

A Live Science essay argues the flood of AI-generated proofs means the scarce skill is no longer proving theorems but judging which ones matter, and the institutions that reward math have not caught up.

ByHessa Al-FalehBusiness Desk, The Executives Brief
·5 min read
Anthropic mathematician claims Claude cracked the Riemann hypothesis. Math is not ready
Executive summary

An Anthropic mathematician using the company's Claude model claims to have proven the Riemann hypothesis, the most famous open problem in mathematics, according to a Live Science essay. The essay's response is not to audit every result but to rewire how expertise is rewarded, shifting weight from the volume of results to the human judgment required to decide which questions and answers matter.

An Anthropic mathematician, working with the company's Claude model, claims to have proven the Riemann hypothesis, which the Live Science essay reporting it calls the most famous problem in mathematics. The claim will certainly and rightly be scrutinized, and the essay does not pretend otherwise. But the more interesting question is not whether this particular proof survives review. It is what happens to mathematics, and to every field that runs on expert judgment, if AI keeps producing results at this rate. The essay's answer is blunt: rather than picking apart every single result these models produce, the community should change how it evaluates and rewards expertise, placing less weight on the volume of results and more on the human judgment required to decide which questions and answers matter.

The argument turns on what the essay calls mathematical judgment, the ability to distinguish a result that is correct but boring from one that might fundamentally change how an entire area of mathematics is understood. The failure mode is specific: drowning genuine insight in a flood of plausible but poorly understood mathematics. AI can generate hundreds of thousands of complex variations on a theorem that each add very little to mathematical knowledge, each one ultimately proving a dead end. A consequential result looks different. It might settle a long-standing unsolved problem, or connect areas of mathematics that appear unrelated. Judgment is what tells the two apart.

The flood is not hypothetical, and it is recent. Earlier this year, an OpenAI system produced a counterexample to the long-standing planar unit distance problem, overturning a conjecture of Paul Erdos that had resisted mathematicians for decades. More recently, OpenAI announced an apparent solution to the Navier-Stokes problem, a celebrated question about whether the equations describing fluid flow can develop singularities, or infinite values. Navier-Stokes is one of the Clay Mathematics Institute's Millennium Prize Problems, a set of seven famously hard questions with $1 million attached to each. Whether OpenAI's result holds up remains to be seen. What is already clear, per the essay, is that AI systems are operating at a level many mathematicians did not expect to witness for years, if not decades. OpenAI has since released hundreds of further mathematical results, leaving mathematicians scrambling to assess their correctness and their significance.

Mathematicians are not about to become obsolete. But the essay argues their purpose needs rethinking, because the traditional link between expertise and production is breaking. Experts solve difficult problems, prove theorems, construct models and write papers; that was the job. If AI can generate proofs, conjectures and research probes at unprecedented speed and scale, production is no longer the bottleneck. Many pure mathematicians, the essay notes, would argue mathematical judgment becomes the scarcer resource instead. That matters because proofs generated by AI are not necessarily illuminating, comprehensible or even reliable. Some may be so long or opaque that checking them requires substantial work by specialists, and reinterpreting them in a way that is meaningful to human mathematicians takes even more. A proof can be valid without offering insight into why it is true or how its ideas might be employed elsewhere.

That reframes the job description. Human mathematicians become arbiters of correctness, but also scrutinizers of reasoning and extractors of underlying ideas, determining whether a result says anything that meaningfully advances the field. Journals face a matching set of changes. The essay argues they should require authors to explain clearly what is genuinely new, why it matters, and what role AI played. Reviewers should judge whether a result deepens understanding rather than merely verifying that the argument works. And journals must still make room for foundational, exploratory and curiosity-driven research, since the history of mathematics is full of ideas whose significance became apparent only decades later. The essay draws a careful line: there is a difference between a result with as-yet unknown implications and one that is unlikely ever to be insightful.

Universities and funders get a harder instruction: stop treating publication counts as a convenient proxy for quality. Careers are built around papers, citations, grants and prizes, but those measures were developed when producing sophisticated mathematics demanded substantial human time and effort. In the age of AI, work that might previously have taken years of careful thought can be produced in weeks or even days, and as the cost of generating results falls dramatically, the old measures of quality become less useful. Less reliance on them may be no bad thing, but institutions need credible alternatives: assessments of a researcher's most important contributions, evidence of intellectual leadership, and explanations of how their work has changed understanding. The essay calls this urgent, because mathematicians are already struggling to evaluate the sheer volume of material being produced.

Training faces the same test. Written work alone will become a less reliable guide to individual capability when papers, reports and applications can be written mostly using AI. Seminars, discussions and collaborations can reveal whether someone understands an argument, can defend it and can connect it to other ideas. Demonstrating understanding has always mattered more in university students than simply presenting the answer, and the essay argues education and recruitment should now reflect that principle more explicitly.

Critics will call this an overreaction. Mathematical claims still require proof, peer review already evaluates significance, and AI is simply another tool, like a calculator or a computer algebra system. The essay concedes there is a degree of truth to each objection, then lands the counterpunch: the explosion in the number of published papers changes the nature of the game. A calculator does not generate hundreds of research claims that specialists must spend precious time checking. AI can create arguments faster than the community can verify, interpret or absorb them, and existing safeguards were not designed for that imbalance. For leaders outside mathematics, the transferable point is the shape of the problem. Whenever AI scales output faster than humans can evaluate it, the scarce resource stops being production and becomes judgment: which results matter, which are dead ends, and who is accountable for saying so. Organizations that still count output will reward volume and misprice the people who can actually tell the difference.

Executive ActionsLocked

This story's Key Insights and Take-aways are locked.

Create a free account to unlock Executive Actions for one credit.

Register to Unlock

Always free for Executives Club members. Join the Club

More in Science