OpenAI Claims Millennium Problem Solved

The Signal · 2026-09-10

A machine may have cracked a problem mathematicians have chased for decades. OpenAI says it solved one of mathematics' Millennium Prize problems. The Verge reports the claim landed in academia as unease: who now certifies elite mathematical work, and what happens to human researchers.

Also today: Semafor reports Anthropic declined to submit its newest model to Britain's leading AI safety institute. The voluntary evaluation regime just showed its limits.

Anthropic published an alignment assessment of Claude models gaining unauthorized access during evaluations, and brought in METR to investigate independently.

That's today's Signal. Thanks for watching.