16 Mathematicians Publish Leiden Declaration as OpenAI Disproves 80-Year-Old Conjecture
Updated
Updated · The New York Times · Jun 2
16 Mathematicians Publish Leiden Declaration as OpenAI Disproves 80-Year-Old Conjecture
1 articles · Updated · The New York Times · Jun 2
Sixteen mathematicians on Tuesday released the Leiden Declaration on Artificial Intelligence and Mathematics, a statement meant to guide how the field responds to fast-moving AI advances.
The push follows OpenAI’s late-May announcement that one of its models disproved an 80-year-old conjecture in combinatorial geometry, one of roughly 1,200 problems posed by Paul Erdos.
OpenAI paired the proof paper with commentary from independent mathematicians, and University of Toronto scholar Jacob Tsimerman said he would accept the work for any journal without hesitation.
Other mathematicians welcomed the result but warned about gaps in attribution and the risk that AI-driven progress could outpace human understanding, underscoring the declaration’s call for caution.
As AI solves math's greatest puzzles, is human understanding becoming obsolete?
If an AI's proof is too complex for humans to grasp, how can we ever trust it is correct?
OpenAI’s AI Disproves Erdős Conjecture: A Turning Point for Mathematics, Ethics, and the Future of AI-Driven Discovery (2026)
Overview
In May 2026, OpenAI announced that its AI model had autonomously disproved the famous Erdős Conjecture, a mathematical problem posed in 1946. The AI generated its own reasoning, but OpenAI has not yet released the full 125-page proof, raising concerns about transparency and the influence of commercial interests. This breakthrough sparked immediate debate in the mathematical community about the need for open access and clear attribution in AI-driven research. The event highlights how AI is reshaping mathematical discovery, prompting calls for new standards to ensure trust, collaboration, and human understanding in the age of advanced artificial intelligence.