Two research teams solved the same open problem with the same AI model, three hours apart
An MIT doctoral student and two professors independently cracked the same question in quantum cryptography using GPT-5.6 Sol Ultra, and posted their papers to arXiv within three hours of each other. It raises an awkward question about what independent discovery means now.
Two groups of researchers who had never coordinated solved the same open problem in quantum cryptography, both using OpenAI’s GPT-5.6 Sol Ultra, and uploaded their papers to the preprint server arXiv three hours apart. Scientific American reported the collision. On one side was Seyoon Ragavan, a doctoral student at MIT. On the other, Prabhanjan Ananth of UC Santa Barbara and Amit Sahai of UCLA.
The problem sits in an area called unclonable encryption, a way of protecting information that leans on a property of quantum physics: certain quantum states cannot be copied, so an encrypted message cannot be quietly duplicated by an eavesdropper. The two groups reached the result by different routes, which is why they are now discussing whether to merge the papers rather than argue about priority.
The quotes are the interesting part. “If someone mentions an open problem, the first thing is to see if GPT solves it,” Ananth said. Ragavan was blunter: “The way I do research now has nothing to do with how I did research two months ago.”
What is behind this. Scientific credit has always run on a rough assumption that hard problems take a long time, so two people arriving at the same answer on the same day is rare enough to be a story. Simultaneous discovery is not new, Newton and Leibniz managed it with calculus, but it used to happen over years, not hours. What changed is that the bottleneck moved. When the hard step was having the idea, timing spread out naturally, because ideas arrive on their own schedule. When the hard step becomes typing a known open problem into a model that many researchers share, the spread collapses. That is a structural change to how credit works, and academia has no mechanism for it yet. Journals award priority to whoever submits first. Three hours is not a meaningful head start on anything. Reactions among mathematicians to this whole shift range from excitement at what is suddenly reachable to something closer to grief about what their work is for.
A caveat that matters: this is a proof in cryptography, a field where a claimed result can be checked line by line and either survives or does not. The verification step is unchanged and still human. Nobody is arguing the model is right because it sounds right.
What this means for you: if you are not in research, the direct effect is nil, and the honest summary is that a model helped two teams do something two teams could already do, faster. The transferable observation is about shared tools generally. When everyone in a field uses the same handful of models, the easy moves get made almost immediately and simultaneously, and whatever advantage is left sits in the questions you think to ask rather than in the answering. That applies to a research group, a consultancy and a marketing team alike. If your edge was speed on a task the model now does, it is gone. If your edge was knowing which problem was worth attacking, it just got more valuable.
Sources
AI agents made research software 60 times faster, and were confidently wrong in ways nobody spotted for weeks
A field report from OpenAI and academic partners documents eight projects where coding agents modernised neglected scientific software. The speedups are real. So are the bugs that only turned up under proper statistical testing.
