On August 11, 2026, Anthropic announced an unreleased model raised the lower bound for the Riemann hypothesis, testing 650 ideas across 60 subagents. The result, confirmed by two in-house mathematicians and formalized in Lean, signals a shift in how AI discovers mathematics.
Key facts
- 31 million output tokens spent on the Riemann hypothesis attempt
- 60 subagents coordinated, with 2 developing key ideas
- 650 different ideas tested by the model
- Lower bound of solutions for Riemann hypothesis increased
- Formalized using open source proof assistant Lean
For over 150 years, the Riemann hypothesis has defied proof, with a $1 million bounty still unclaimed. According to TechCrunch, Anthropic's unreleased model didn't solve it but made significant progress by increasing the lower bound of solutions for which the hypothesis holds. The work, announced Monday, is the latest in a string of AI-driven mathematical breakthroughs.
How the model worked
The progress came from an unusual setup. An Anthropic staff member without significant mathematical training prompted the model to "take a real stab" at proving the hypothesis, then left it to coordinate the task over a day and a half. The model tested 650 different ideas, coordinating across 60 subagents and spending 31 million output tokens. A footnote to the paper details the division: two subagents developed key ideas, 13 contributed ideas, 30 attempted but failed, 13 validated arguments, and two wrote the initial paper.
The finding was confirmed by two of Anthropic's in-house mathematicians and formalized using the open source proof assistant Lean. This formalization is critical — it provides a machine-checkable verification, addressing concerns about AI-generated proofs' reliability.
Broader context and controversy
The result sits within a wave of AI mathematical discoveries. OpenAI recently released 10 major results proved by its internal "Astra" model, while Anthropic's separate effort disproved the longstanding Jacobian conjecture. According to TechCrunch, a group of prominent mathematicians signed a June declaration warning that AI could undermine the field's standard that proofs be "attributable to specific authors who take credit for their discovery and assume responsibility for their correctness."
Fields Medal winner Timothy Gowers responded in a blog post, questioning whether AI's influence might change mathematics in a more complex, positive way. The field remains split. The Anthropic result, with its subagent coordination and Lean verification, suggests a future where AI-orchestrated proofs become more common — but the attribution problem persists.
Key Takeaways
- Anthropic's unreleased model raised the lower bound for the Riemann hypothesis, testing 650 ideas with 60 subagents, confirmed by mathematicians and Lean.
- This signals AI's growing role in mathematical discovery.
What to watch
Watch for the full paper's release and whether Anthropic discloses the model's name and training details. Also track OpenAI's response to this result, and whether the mathematical community adopts Lean-based verification as a standard for AI-generated proofs.
Source: techcrunch.com
