Anthropic’s Unreleased AI Model Makes Progress on the Riemann Hypothesis

Anthropic has not solved the Riemann hypothesis, one of mathematics’ major unsolved problems for more than 150 years. TechCrunch reports that an as-yet-unreleased Anthropic model made significant progress by increasing the lower bound of solutions for which the hypothesis holds true. The problem remains open, and the $1 million bounty for a general proof remains unclaimed.

The work began when an Anthropic staff member without significant mathematical training prompted the model to “take a real stab” at proving the hypothesis. The model then coordinated work over the following day and a half, testing 650 different ideas across 60 subagents and producing 31 million output tokens. According to a footnote cited by TechCrunch, some subagents developed key ideas, others contributed or validated arguments, and two helped write the initial paper.

The result was confirmed by two Anthropic in-house mathematicians and formalized using Lean, an open source proof assistant. That verification step is important because progress in high-level math depends not just on plausible ideas, but on arguments that can be checked for correctness. The article frames the result as part of a broader wave of LLM-assisted mathematical work rather than a final proof of the hypothesis.

The report highlights both excitement and concern inside the mathematics community as AI systems contribute to more advanced research. A public declaration signed in June warned that AI could challenge norms around attribution, responsibility, and proof authorship. Fields Medal winner Timothy Gowers, cited by TechCrunch, suggested the impact could be more complex and potentially positive, comparing unnamed theorems to unnamed stars.