OpenAI’s latest model, GPT-5.6 Sol Ultra, has produced a proof of the Cycle Double Cover Conjecture—a graph theory problem that stumped mathematicians for 50 years. The AI did it in under an hour by running 64 subagents in parallel, each trying variations until something clicked. Mathematician Thomas Bloom says the proof is elegant and could have been found in the 1980s, but humans likely gave up too early when the obvious approach failed. AI doesn’t get discouraged; it just keeps tweaking. However, Bloom also notes the AI’s paper lacks citations for key ideas that trace back to a 1983 paper—a common issue with AI-generated work. The proof is clever, but the AI likely pulled ideas from existing literature without credit. A full human verification is still pending. Why it matters: This shows AI’s raw persistence can solve problems humans abandon, but also highlights a dangerous blind spot around attribution that could undermine trust in AI research.