An AI model just did something a working scientist couldn’t do for three years. According to OpenAI, GPT-5 Pro helped immunologist Derya Unutmaz crack an immunology mystery that had stumped his lab, producing fresh insight into how T cells behave. OpenAI shared the case as an example of its newest model contributing to live biomedical research, not a textbook recap.
What stands out here is the timeline. This wasn’t a question with a known answer sitting in a paper somewhere. Unutmaz had been circling the problem for three years. GPT-5 Pro reportedly helped surface the connection that moved it forward.
What actually happened
T cells are the immune system’s frontline soldiers. They hunt down infected and cancerous cells, and they’re central to how the body either defends itself or, when things go wrong, attacks its own tissue. Understanding why they behave the way they do is one of the hardest open problems in immunology.
Per OpenAI, the model gave Unutmaz insights into T cell behavior that pointed toward an explanation for his long-running question. The company frames the breakthrough as relevant to two big research fronts:
- Cancer research, where T cells can be trained to attack tumors
- Autoimmune research, where T cells mistakenly turn on healthy cells
Progress on either front has real stakes for treatments down the line.
Why this matters
We’ve heard a lot of claims about AI “transforming science.” Most of them stay vague. This one is specific: a named researcher, a defined problem, a concrete result.
That’s the shift worth watching. Earlier models were useful for summarizing literature, drafting code, or brainstorming hypotheses. Helping a domain expert close out a three-year question is a different level of contribution. It moves the model from research assistant toward research collaborator.
The key detail is that Unutmaz is an expert in his field. GPT-5 Pro didn’t replace his judgment. It gave him a thread he hadn’t connected, and he had the knowledge to recognize it mattered and run with it. That pairing, deep human expertise plus a model that can reason across huge amounts of information, is where the real value shows up.
The bigger picture
This lands during a broader push to point frontier AI at hard science. Labs across the industry are testing whether these models can help with drug discovery, protein work, and complex biological reasoning. OpenAI publishing a named immunology case is part of making that argument with evidence rather than promises.
A few things to keep in perspective:
- One case isn’t a pattern. A single breakthrough is encouraging, not proof that this works on demand across every lab and every problem.
- Verification still belongs to scientists. Any insight a model surfaces has to be tested at the bench. The model proposes; the experiments decide.
- Expertise is the multiplier. The result came from a specialist who knew how to use the tool, not from the tool alone.
What to expect next
If you work in or near research, the practical takeaway is straightforward: it’s worth putting your stubborn, long-standing questions in front of a capable model. Not because it’ll hand you the answer, but because it might connect dots you’ve been staring past. The cost of trying is low. The upside, as this case shows, can be a problem you’d written off.
Expect OpenAI and its competitors to surface more stories like this one as they compete to prove their models earn a seat in serious scientific work. Watch for whether these results start showing up in peer-reviewed papers, which is where claims like this get their real test.
For the full account of how GPT-5 Pro helped Unutmaz reach the breakthrough, the original write-up from OpenAI has the details.