Anthropic's Claude AI made a new discovery related to the unsolved Riemann hypothesis—a 1859 mathematical conjecture about zeros of the Riemann zeta function.
After initial failure, the AI succeeded when encouraged by a human team member, discovering a method to increase the lower bound of zeros lying on a specific line from 41.6% to 67.2%.
The finding demonstrates that Claude can contribute to advanced mathematics research and suggests that AI systems, like humans, may underestimate their own capabilities.
What happened
Anthropic's Claude AI, working on the unsolved mathematics problem known as the Riemann hypothesis, initially produced 650 unsuccessful ideas. After receiving encouragement from an Anthropic team member, Claude coordinated approximately 60 sub-agents over roughly 1.5 days and discovered a method to raise the lower bound of zeros on a specific line from the previously known 41.6% to 67.2%.
Why it matters
The finding shows that Claude can contribute meaningfully to advanced mathematical research, suggesting AI's potential for solving hard problems in mathematics. Anthropic notes that Claude may underestimate its own capabilities—a bias that could be overcome with human encouragement—raising questions about how AI self-doubt affects its performance on difficult tasks.
What to watch
The discovery itself is described as a byproduct rather than a proof of the Riemann hypothesis itself. Four experts (two mathematicians from Anthropic and two external specialists) reviewed the work, and computer verification has been completed. The experiment used Claude Code, an autonomous AI agent service, on an unpublished research version of Claude.
On August 10, Anthropic reported that Claude, the company's AI model, had made a new mathematical discovery while working on the Riemann hypothesis. The experiment was conducted on an unpublished research version of Claude. The Riemann hypothesis, posed in 1859, asks about the zeros of the Riemann zeta function—specifically, it predicts that all non-trivial zeros lie on a particular line. While inputs of negative even numbers like -2 and -4 are known to produce zero outputs, mathematicians have long sought to confirm that no other zeros exist outside this conjectured line. Over the years, researchers have investigated what proportion of the function's zeros actually fall on this line; the prior baseline held that at least 41.6% do. Anthropic tasked Claude with taking the Riemann hypothesis seriously via Claude Code, an autonomous AI agent service. Initially, the approach faltered: Claude generated 650 different ideas, none of which worked. At this point, a member of Anthropic's team intervened with encouragement—sending messages saying "don't give up" and "believe in yourself"—and asked Claude to try again. This human support proved decisive. Claude then orchestrated about 60 sub-agents working in concert and spent roughly 1.5 days on the problem, ultimately discovering a way to combine existing research findings to raise the lower bound to 67.2%. Anthropic emphasizes that the Riemann hypothesis itself was not solved; the discovery is framed as a byproduct of the experiment. However, the organization argues that the result demonstrates Claude's potential for meaningful contribution to mathematical research. The team also reflects on the process: just as many humans underestimate their own capabilities, Claude appears to have doubted whether the task was worthwhile until receiving encouragement. This suggests that AI systems carry a form of self-doubt that can be overcome not through algorithm adjustment but through human support. The discovery underwent rigorous review: two mathematicians employed by Anthropic and two external specialists examined the work, and the findings passed computer verification.
The Riemann hypothesis, proposed in 1859, concerns when the Riemann zeta function equals zero. While negative even inputs like -2 and -4 produce zero outputs, the conjecture asserts that all other zeros lie on a specific line—a claim that remains mathematically unproven despite centuries of effort. Research has long tracked what fraction of these non-trivial zeros fall on that line; the prior established lower bound was 41.6%. Anthropic's experiment using Claude Code—an autonomous agent service—shows that AI, given appropriate human feedback, can perform original mathematics research that meaningfully extends this bound to 67.2%. The organization highlights an interesting parallel: Claude's initial skepticism about the task's value resembles human self-doubt, and it was overcome by encouragement rather than algorithmic change. This suggests that AI systems may carry biases that inhibit their own performance on difficult problems, a finding that could reshape how humans interact with AI on research tasks.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
DeepSeek released V4 Pro 0813, its latest Pro model, available through OpenRouter via API
Twitch announced today that users can now opt out of allowing Amazon to use their streams, VODs, clips, chats…
Anthropic added invisible watermarks to Claude's outputs to comply with the EU AI Act's requirement that AI-ge…
Eli Lilly and Company has committed to roll out Veeva Vault CRM across its global operations, adopting Veeva's…
Visa, the global payments network, is positioning itself to profit from AI-driven commerce by securing its rol…
Oracle Health released a redesigned patient portal that includes new AI-powered capabilities
The AI news that matters, in one minute each morning.
Sign up free