Two teams used the same AI model and got the same proof
Two research groups gave GPT-5.6 Sol Ultra the same open problem and filed proofs 3 hours apart. What that means when your competitor runs the same model.
Two research groups pointed the same AI model at the same open problem and landed the same answer three hours and eighteen minutes apart. Neither knew the other was working on it. That collision is the most useful thing that happened in AI this week, and it has nothing to do with the math.
What actually happened
Per Scientific American, UC Santa Barbara's Prabhanjan Ananth and UCLA's Amit Sahai posted a paper to arXiv at 10:35 A.M. PDT. MIT PhD student Seyoon Ragavan posted his 3 hours and 18 minutes later. Both papers resolved the same open question in unclonable encryption — whether an efficient version exists without extra security assumptions — a problem that had sat open since Broadbent and Lord framed it in 2019. Both credited OpenAI's GPT-5.6 Sol Ultra with the core ideas.
The workflows were not identical. Ragavan drove the model in successive two-hour stretches, redirecting it when it wandered, then cleaned up and reorganized the proof it produced. Ananth and Sahai ran it through a custom UCLA harness built to make the model critique its own attempts. Different operators, different scaffolding, same result, same morning.
Neither paper is peer-reviewed yet. The humans verified the output and took responsibility for the claims — that part didn't get automated.
Why the same AI model is a problem for your business
Everyone reading this has access to the same frontier models as their competitors. Same weights, same API, same price list. If two of the sharpest cryptographers alive can be beaten to their own result by three hours because a stranger asked the same question, your "AI-powered" differentiator has a shelf life measured in weeks.
What actually survives: the question you know to ask, the proprietary context you can feed the model, and the verification step you own. Ananth's line — that the first move on any open problem now is to see whether the model solves it — is the whole game. The model is a commodity. The prompt informed by ten years of running your business is not.
Practically: stop building product around a capability the model gives everyone for free. Build around the data only you have — your order history, your support transcripts, your service records — and the workflow that turns model output into something a customer can trust. Both papers still needed a human to check the proof. Yours will too.
Key takeaways
- Two independent teams used GPT-5.6 Sol Ultra on the same unclonable-encryption problem and filed papers 3h18m apart
- Different workflows — manual redirection vs. a custom self-critique harness — produced the same core result
- Model access is not a moat; your competitors have the identical model at the identical price
- Defensible AI work sits in proprietary data, domain-specific questions, and the verification step you own
We build AI systems around the data your competitors don't have. If your AI project is a wrapper anyone could rebuild in a weekend, tell us what you're trying to do and we'll show you where the actual leverage is. See what we've shipped.
Sources: Scientific American.
- #ai-research
- #competitive-advantage
- #ai-workflow
- #model-strategy
- #differentiation
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
An AI agent attacked 460 systems. Look at what it targeted.
Unit 42 documented an autonomous AI attack campaign against 460+ targets. Four of seven exploit tracks hit self-hosted automation and AI tooling — patch that first.
Read itQwen3.8-Max goes wide, open weights land next week
Alibaba opened Qwen3.8-Max to global developers ahead of an open-weights release. At 2.4T parameters, open weights isn't self-hosting — here's what it actually buys you.
Read it