MIT doctoral candidate Seyoon Ragavan and a two-professor team, Prabhanjan Ananth of UC Santa Barbara and Amit Sahai of UCLA, independently cracked the same open problem in quantum cryptography. Both sides used OpenAI’s GPT-5.6 Sol Ultra. Both filed papers to arXiv within three hours of each other, and both are now discussing whether to combine the work into one submission, The Decoder’s Matthias Bastian reported on August 3, citing original reporting from Scientific American.
Simultaneous discovery is not new. Newton and Leibniz split credit for calculus. Darwin published on natural selection only after learning Alfred Russel Wallace had reached the same conclusion independently. Bell beat Gray to the patent office by hours. Historians treat these collisions as evidence that ideas arrive when the surrounding knowledge makes them reachable, not as freak coincidences.
What changed this time is the mechanism, not the outcome. Newton and Wallace converged because two minds independently absorbed the same accumulating evidence. Ragavan, Ananth, and Sahai converged because they queried the same model. When the search for a proof runs through a shared oracle rather than through separately built intuition, the distance between two researchers’ next move shrinks. Near-simultaneous results stop being a coincidence worth remarking on and start being the arithmetic outcome of everyone drawing from one pool of reasoning.
The researchers describe the shift in blunt terms. “If someone mentions an open problem, the first thing is to see if GPT solves it,” Ananth said. Ragavan put it more starkly: “The way I do research now has nothing to do with how I did research two months ago.”
That should worry anyone who still treats being first as the currency of a career. If two labs can produce adjacent proofs three hours apart, credit stops tracking insight and starts tracking who hit submit sooner, a metric that rewards speed of typing over speed of thought. Peer review gets stranger too: two independent submissions that share a hidden coauthor in the form of the same model are not independent verification of an idea, they are two transcriptions of one machine’s reasoning wearing different names.
One detail cuts against the cleanest version of this argument: Ragavan and the Ananth-Sahai team built their proofs on separate methods within unclonable encryption, the quantum-based technique both papers address. If the model were fully collapsing the search space, both proofs would look more alike than they do. The truer read is that GPT-5.6 narrowed the distance between researchers without erasing it entirely, which is exactly the transition period worth watching before it fully closes.
Today’s related announcement of ten new proofs from an internal OpenAI model is the other half of this story: not what the models can find, but what it does to the people racing them there.
Reporting by Matthias Bastian for The Decoder on August 3, 2026, drew on original coverage from Scientific American.
