Two Research Teams Independently Solve Same Quantum Crypto Problem Using GPT-5.6, Three Hours Apart

Two separate research teams solved the same open problem in quantum cryptography using OpenAI's GPT-5.6 Sol Ultra, submitting their papers to arXiv.org roughly three hours apart, according to a report by Scientific American.

MIT PhD student Seyoon Ragavan worked independently on the problem, while UC Santa Barbara professor Prabhanjan Ananth and UCLA professor Amit Sahai tackled it as a separate team. Both parties used the same AI model but arrived at different technical approaches to "unclonable encryption," a cryptographic method that relies on quantum mechanical properties to prevent data from being copied. The two groups are now discussing whether to merge their papers.

Same tool, same problem, same day

The near-simultaneous submissions highlight a shift in how theoretical research is being conducted. According to Ananth, checking whether GPT can solve an open problem has become a standard first step: "If someone mentions an open problem, the first thing is to see if GPT solves it."

Ragavan described an even more fundamental change in his workflow. "The way I do research now has nothing to do with how I did research two months ago," he told Scientific American.

Neither Scientific American nor the researchers disclosed specific technical details about how GPT-5.6 Sol Ultra was applied to the proofs, what prompting strategies were used, or how much of the final papers were AI-generated versus human-verified. OpenAI has not issued a statement on the case, and pricing, context window, and benchmark specifications for GPT-5.6 Sol Ultra were not included in the report.

A question of originality

The incident raises a question that extends beyond cryptography: when multiple researchers have access to the same frontier model, and that model can generate similar solution paths, what qualifies as independent discovery? Two people arriving at the same proof using the same tool within hours of each other blurs a line that peer review and citation norms were built around — the assumption that separate research groups working in isolation represent genuinely separate intellectual effort.

Mathematics appears to be an early testing ground for this dynamic. The field's formal, verifiable nature makes it well-suited to AI-assisted proof generation, and reactions among mathematicians described in the report range from enthusiasm about new problem-solving capacity to unease about what the change means for professional identity and the value of individual insight.

What this means

This case is presented by Scientific American as an anecdote, not a controlled study — there's no independent verification of how much the AI contributed to each proof versus how much was human refinement. Still, the pattern is worth watching: as frontier models become common lab equipment for theoretical researchers, simultaneous discovery may become routine rather than remarkable. That could accelerate fields like cryptography and pure math, but it also pressures existing norms around credit, priority claims, and what peer review is meant to certify. Expect journals and funding bodies to start asking authors to disclose AI tool use in proofs, similar to current disclosure requirements for AI-assisted writing.