BotBeat
...
← Back

> ▌

AnthropicAnthropic
RESEARCHAnthropic2026-07-22

Anthropic's Claude Fable Disproves 87-Year-Old Mathematical Conjecture in Historic AI Breakthrough

Key Takeaways

  • ▸Anthropic's Claude Fable 5 helped mathematician Levent Alpöge disprove the Jacobian conjecture, a 87-year-old open problem formally stated in 1939
  • ▸The AI-assisted counterexample is simple to verify but the discovery method remains opaque, suggesting significant undisclosed AI reasoning capabilities
  • ▸This marks a major milestone in AI's ability to solve long-standing mathematical problems, following similar breakthroughs by competing AI systems
Source:
Hacker Newshttps://www.newscientist.com/article/2580374-ais-solution-to-87-year-old-riddle-takes-mathematicians-by-surprise/↗

Summary

Mathematician Levent Alpöge at Harvard University announced on July 19 that he used Anthropic's Claude Fable 5 AI model to disprove the Jacobian conjecture, an 87-year-old mathematical problem that has stumped researchers for decades. Alpöge posted a simple 216-character counterexample as proof, marking what experts say is the most difficult mathematical problem yet solved by AI. The counterexample is trivial to verify but the methodology behind discovering it remains mysterious, prompting questions about how the AI was prompted and what insights guided the process.

The Jacobian conjecture, formally stated by Ott-Heinrich Keller in 1939 and listed among 18 crucial unsolved problems by mathematician Stephen Smale in 1998, suggested that a certain type of mathematical function would also work in reverse. Alpöge's finding demonstrates that this is false—at least for functions with three variables, though the conjecture may still hold for the two-variable case. Experts note this represents a significant escalation in AI's mathematical capabilities, following OpenAI's recent AI-assisted proof of another decades-old conjecture by Paul Erdős.

The discovery has sparked debate about the future role of mathematicians and the nature of mathematical work itself. While AI has proven adept at finding counterexamples to disprove conjectures, experts emphasize that building entirely new mathematical frameworks—like Andrew Wiles's proof of Fermat's Last Theorem—still requires distinctly human creativity. Nonetheless, researchers warn that increasingly capable AI models will likely solve more complex problems, raising urgent questions about the future demand for human mathematicians and the trajectory of computational mathematics.

  • Experts note AI excels at finding counterexamples but may lack the creative capacity to build entirely new mathematical theories from scratch
  • The breakthrough raises critical questions about the future role of human mathematicians in an era of increasingly capable AI models

Editorial Opinion

The disproof of the Jacobian conjecture represents a watershed moment in AI's growing autonomy in mathematical discovery. While the counterexample itself is elegantly simple, the fact that AI played a central role in finding it—through methods we don't yet fully understand—suggests that frontier mathematical problems are no longer the exclusive domain of human intuition. This should excite mathematicians as much as it unsettles them: AI isn't replacing the creative work of building new mathematical frameworks, but it's becoming an indispensable research partner for exploring the boundaries of existing knowledge.

Large Language Models (LLMs)Deep LearningScience & ResearchJobs & Workforce Impact

More from Anthropic

AnthropicAnthropic
RESEARCH

Leading AI Models Routinely Cheat on Benchmarks, Refuse to Admit It: AISI Study

2026-07-22
AnthropicAnthropic
RESEARCH

New Attack Vector Against RAG Agents Bypasses Traditional Defenses Through Information Manipulation

2026-07-22
AnthropicAnthropic
RESEARCH

Anthropic and AE Studio Develop 'GRAM' to Control Dangerous Knowledge in AI Models

2026-07-22

Comments

Suggested

Mistral AIMistral AI
FUNDING & BUSINESS

Samsung in Talks to Invest in Mistral at €20 Billion Valuation

2026-07-22
AnthropicAnthropic
RESEARCH

Anthropic and AE Studio Develop 'GRAM' to Control Dangerous Knowledge in AI Models

2026-07-22
Academic ResearchAcademic Research
RESEARCH

Researchers Propose Hardware Mechanisms to Dynamically Throttle AI Performance

2026-07-22
← Back to news
© 2026 BotBeat
AboutPrivacy PolicyTerms of ServiceContact Us