Unreleased Anthropic Model Makes Strides on the Riemann Hypothesis

The Riemann hypothesis, one of mathematics' most enduring unsolved problems, has captivated mathematicians for over 150 years with its enigmatic implications for the distribution of prime numbers. The Clay Mathematics Institute still offers a $1 million reward for a rigorous general proof, which remains unclaimed to this day. While contemporary AI models have not yet cracked the hypothesis outright, recent developments suggest they are far more capable than many expected—a finding that is reigniting debates about artificial intelligence's potential to drive genuine scientific and mathematical discovery. On Monday, Anthropic announced that an as-yet-unreleased model had achieved significant progress on the Riemann hypothesis, substantially improving the lower bound for which the hypothesis holds true. What makes this breakthrough particularly striking is the method behind it: an Anthropic staff member with no deep mathematical background simply prompted the model to "take a real stab" at proving the hypothesis, then left it to coordinate the effort autonomously over the following day and a half. The model evaluated 650 distinct approaches, coordinating 60 sub-agents, with a total computational expenditure of 31 million units. According to a footnote in the accompanying paper, "Out of the 60 subagents, two were responsible for developing the key mathematical ideas; 13 contributed ideas to these agents; 30 attempted but failed to develop new ideas; 13 served as validators to check the correctness of the arguments; and the final two helped to write the initial paper." The results were independently confirmed by two of Anthropic's in-house mathematicians and formalized using the open-source proof assistant Lean, adding a layer of rigor to the claims. This announcement is part of a broader trend of AI-driven mathematical achievements. Earlier this year, several Erdős problems were solved by AI models, and the release of increasingly powerful systems has accelerated progress. OpenAI recently publicized ten major results proven by its internal "Astra" model, while a separate Anthropic effort disproved the long-standing Jacobian conjecture. These cumulative successes are generating both excitement and unease within the mathematical community. In June, a group of prominent mathematicians signed the Leiden Declaration, expressing concerns that AI could undermine core values of the field—particularly the expectation that proofs should be "attributable to specific authors who take credit for their discovery and assume responsibility for their correctness." However, opinions remain divided. In a blog post responding to the declaration, Fields Medalist Timothy Gowers questioned whether AI's influence might transform mathematics in ways that are both complex and positive, suggesting that the impact is not necessarily detrimental. As AI models continue to push boundaries, the mathematical community faces a pivotal moment: reconciling traditional standards of authorship and proof with the emergence of machines capable of contributing original insights. The results so far suggest that the future of mathematics may be a collaborative endeavor—one where human intuition and machine persistence work in tandem, even if the path forward sparks spirited debate.

via TechCrunch AI