GPT-5.6 helps two teams post same crypto proof hours apart

Two independent teams posted proofs of the same open problem in unclonable encryption to arXiv three hours and eighteen minutes apart, and both credited OpenAI's GPT-5.6 Sol Ultra as a working collaborator, according to Scientific American. MIT PhD student Seyoon Ragavan uploaded his version three hours and eighteen minutes after UCSB's Prabhanjan Ananth and UCLA's Amit Sahai filed theirs at 10:35 A.M. PDT.

The problem sits inside quantum cryptography and builds on a 2019 framework from Anne Broadbent and Sébastien Lord. The two teams took different routes to the same result. Ragavan worked conversationally, letting the model run in two-hour stretches and course-correcting between them. Ananth and Sahai fed the problem to a bespoke UCLA system designed to help models pursue and critique possible solutions. That two teams arrived at the same result, on essentially the same day, with essentially the same tool, is the part worth sitting with.

The quotes from the researchers are what make this hard to file as hype. Ragavan told the magazine that "the way I do research now has nothing to do with how I did research two months ago" and, on the timeline, said "it's like two weeks and a day since this idea even formed." Ananth put the new default plainly: "if someone mentions an open problem, the first thing is to see if GPT solves it." Broadbent, whose 2019 work with Lord the proofs build on, said "in hindsight, this was an obvious thing to look into" and told Scientific American she has "a lot of questions about haves and have-nots."

The honest caveat is that this is one story, one problem, one subfield, sourced from the researchers involved in the moment. The reporting doesn't lay out how much of the proof text was model-authored versus human-directed, what peer review of these arXiv preprints will look like, or how UCLA's bespoke critic system differs technically from a vanilla GPT session. Take the specifics as reported, not settled.

What is worth watching is who this rewards. Research groups with in-house tooling like UCLA's, or the budget to iterate against a frontier model at length, now have a real time-to-proof advantage over the backlog of open conjectures. Ananth said he is "happy that I am past being a student" and worries "for the current crop." That is probably the right instinct to hold with this one.