Anthropic's Unreleased AI Model Makes Progress on Riemann Hypothesis

Anthropic announced Monday that an unreleased internal artificial intelligence model has made substantial progress on the Riemann hypothesis, a 150-year-old mathematical conjecture concerning the distribution of prime numbers and currently subject to a 1 million Millennium Prize. The development marks a significant milestone in automated scientific discovery and has intensified ongoing debates regarding the role of machine learning in advanced mathematics. The breakthrough emerged when an Anthropic engineer, lacking specialized mathematical training, instructed the model to attempt a formal proof. Left to operate autonomously for thirty-six hours, the system coordinated a workforce of sixty sub-agents, evaluated six hundred fifty distinct theoretical approaches, and consumed approximately thirty-one million units of computational resources. Project documentation indicates that two primary agents generated the core mathematical insights, supported by thirteen idea contributors, thirty exploratory agents, thirteen verification units, and two composition agents. The resulting framework successfully elevated the lower bound of solutions for which the hypothesis holds true. The model output was subsequently reviewed and verified by two Anthropic mathematicians and formalized using Lean, an open-source proof assistant designed to guarantee mathematical rigor. This achievement aligns with a rapid acceleration in AI capabilities across mathematical challenges. Over the past year, large language models have resolved multiple Erdos problems, while Anthropic previously helped disprove the Jacobian conjecture. Similarly, OpenAI recently disclosed ten major proofs generated by its internal Astra model, demonstrating a clear industry shift toward autonomous theorem generation. The expanding capabilities of AI in mathematics have fractured academic consensus regarding future research paradigms. In June, a coalition of prominent mathematicians published a public declaration expressing concern that autonomous proof generation could erode foundational academic values, particularly the requirement that discoveries remain attributable to individual researchers who assume responsibility for their validity. Conversely, Fields Medal recipient Timothy Gowers challenged this stance, suggesting that decoupling theorems from human authorship may not fundamentally diminish the discipline. He compared the phenomenon to celestial naming conventions, arguing that mathematical truth remains valid regardless of whether it originates from human intuition or machine computation. As AI systems transition from pattern recognition to active hypothesis generation, the mathematical community faces a critical inflection point regarding validation standards and academic authorship. Anthropic's demonstration underscores both the unprecedented analytical capacity of modern language models and the urgent need for structured guidelines governing AI-assisted scientific breakthroughs. The Riemann hypothesis remains unproven, but the trajectory of human-machine collaboration in mathematics has decisively shifted toward hybrid research methodologies.