Two labs handed AI the same cryptography puzzle. It answered both before lunch.

On a single morning in July, two research groups on opposite coasts asked the same AI for help on the same obscure cryptography problem. Within three hours, both had submitted their own arXiv preprints, papers posted to arXiv, the open server where physicists and computer scientists post work before peer review, proving the same result. The question had been posed earlier that month at the Simons Institute for the Theory of Computing in Berkeley, and it sat in a specialized corner of quantum cryptography called unclonable encryption, a family of schemes that exploit the no-copy rule of quantum information to make encoded messages physically impossible to duplicate.

Both papers credit OpenAI's GPT-5.6 Sol Ultra with surfacing the core ideas behind the proofs and the constructions that made them work. Neither has been peer-reviewed. Both sit in the same blind spot: the humans still picked the question, still checked the steps, and still wrote the paper. What the model did was the part in the middle.

The first paper is Ragavan. The second is co-authored Ananth. Ananth's doctoral student Yao-Ting Lin was in a meeting with his adviser when word of Ragavan's parallel result came in. Lin opened the email, then opened Ragavan's paper. "We are definitely living in strange times," he replied.

The simultaneity is the part the field has to absorb. Both Ragavan and Sahai heard the open question at Simons in early July 2026, then pursued it separately, using the same model through different workflows. They did not coordinate. They did not compare notes. The result is the same. That outcome is a useful test case, not a fluke, and the pressure it puts on the credit system is the real signal.

"If someone mentions an open problem, the first thing is to see if GPT solves it," Ananth told Scientific American. "A lot of problems that we didn't know how to solve are going to get solved in the very near future." That is honest, and it is also the cultural shift the field is now negotiating. Scientific American disclosed that its reporter works in the same lab as the Simons speaker.

Unclonable encryption is a specialization of quantum cryptography that turns the laws of physics into a security guarantee. The no-cloning theorem says a quantum state cannot be perfectly copied, so a ciphertext built from one is, in a precise sense, uncopyable. Schemes that exploit this have been a niche research target for years. They are not, in the language of the field, easy. The fact that the same model produced the core ideas for two proofs on the same morning says less about the model's reach than it does about how thin the bottleneck has become in some parts of theoretical computer science.

The credit question is older than the model. Simultaneous discovery has happened in physics, in mathematics, in biology, and the field has existing norms for handling it. Those norms tend to hinge on who arrived at the result independently. When both groups route through the same underlying engine, "independence" starts to bend. The work of designing new norms, what counts as the discovery, who gets authorship, how verification is documented when the same model is in both pipelines, is the part that has not been written yet.

Humans still pick which questions are worth asking, and they still have to verify the steps. A proof is a proof because a reader can walk through it, and that reader is still a person. What the model adds is the search itself: candidate constructions, rough drafts, the kind of work that used to take a research team weeks. The bottleneck moves up, from "find the trick" to "frame the question" and "check the answer." That is a real change, and it is one the field can design for, if it chooses to.

What does not change is the simple fact at the center of this story: two papers, three hours apart, on the same obscure problem, in the same corner of the field, with the same model credited for the core work. The work of figuring out what that means for credit, for authorship, and for what gets called a discovery is just beginning.