Two Teams Solved the Same Crypto Problem Hours Apart, but Only One Confirmed Using AI
An MIT student working alone credits GPT-5.6 Sol Ultra directly in his paper. A UCSB-UCLA team reached the same unsolved cryptography proof, reportedly using the same model through a different setup. Their papers landed on arXiv three hours and eighteen minutes apart.
Two teams reached the same crypto proof hours apart. Only one confirms it used GPT-5.6.
MIT PhD student Seyoon Ragavan, working alone, used OpenAI's GPT-5.6 Sol Ultra to solve an open problem in quantum cryptography, an efficient unclonable encryption scheme, according to his own paper. A team of Prabhanjan Ananth at UC Santa Barbara and Amit Sahai at UCLA reached the same solution using the same model, according to Scientific American's reporting, though their own paper doesn't mention AI use at all. The two papers landed on arXiv 3 hours and 18 minutes apart on July 23, 2026.
How close the timing actually was
Ananth and Sahai submitted first, at 17:35:44 UTC. Ragavan submitted at 20:53:24 UTC, the same day. Both teams had heard the same open problem posed at a Simons Institute event at UC Berkeley earlier that month. Secondary reporting describes the two proofs as essentially identical except for one lemma's proof technique.
Not quite the same method
Ragavan's own paper credits GPT-5.6 Sol Ultra directly in its text: the model found the proof in an extended conversation with him and drafted a preliminary version of the paper, with Ragavan stating he is fully accountable for its correctness. Ananth and Sahai's paper doesn't mention AI use at all in its own text. Per Scientific American's reporting, they worked through a custom system built at UCLA designed to have the model pursue and critique its own candidate solutions, a more structured setup than Ragavan's direct back-and-forth.
What it means for research
Ananth, on record: 'Now the general mentality is: if someone mentions an open problem, the first thing is to see if GPT solves it.' The specific timing here, two independent teams landing on the same solution within hours without knowing about each other, is a real, checkable data point for how much AI is compressing the time between a problem being stated and a proof existing.
Why a build studio cares
This isn't really about cryptography. It's about what happens to credit, priority, and peer review once the bottleneck on solving a hard problem stops being human effort and starts being who thought to ask the model first.
Next step: read Scientific American's coverage or the papers themselves on arXiv: Ragavan and Ananth and Sahai. If AI-assisted research workflows are relevant to your own work, write to us at hello@gattyworks.com.