GPT-5 Pro Helps an Immunology Lab Crack a Years-Long T-Cell Mystery

When a Model Becomes a Genuine Research Partner

Every now and then a story lands that captures exactly why so many of us are optimistic about AI in the sciences. On June 24, 2026, OpenAI shared a case study in which GPT-5 Pro helped an immunology lab explain a puzzle it had been circling for roughly three years. As someone who reads a lot of these claims with a careful eye, I found this one worth slowing down for — because the details hold up.

The researcher is Derya Unutmaz, MD, of The Jackson Laboratory for Genomic Medicine. His team had been trying to understand how glucose influences the way T cells develop and specialize — a question with real implications for cancer and autoimmune research. The lab had data and intuition, but the underlying mechanism stayed stubbornly out of reach.

What GPT-5 Pro Actually Contributed

Let me separate confirmed reporting from interpretation, as I always try to do. According to the case study, GPT-5 Pro proposed a mechanistic explanation the lab had not landed on: that deoxyglucose blocks the protein IL-2, which in turn helps explain why it nudges T cells toward an inflammatory "Th17" path. That is a specific, testable claim about *why* something happens, not a vague summary of existing literature.

The Detail That Rules Out Simple Lookup

Here is the part that impressed me most. The model also correctly predicted the outcome of unpublished experiments involving CD8+ T cells attacking lymphoma cells. Because those results were not in any training data, this wasn't a case of retrieving a known answer — it points to the model reasoning over the biology rather than echoing a paper. Unutmaz described the effect on his lab vividly, saying the tool changed "the tempo at which expert immunology can move."

Augmentation, Not Replacement

I want to be precise about the framing, because it matters. This is a story about AI augmenting expert scientists, not standing in for them. The hypotheses still had to be evaluated by a domain expert who knew which questions were worth asking and how to validate an answer at the bench. The model accelerated the work; the human judgment anchored it. That partnership — fast machine reasoning paired with careful human verification — is, in my view, the most promising shape for AI in scientific research.

Why This Direction Is Encouraging

Drug discovery and immunology move at the pace of hypotheses tested. Anything that helps skilled researchers generate better hypotheses faster compounds over time. A single mechanism explained today can reshape which experiments a lab runs next month. Multiply that across thousands of labs and you start to see why tools like GPT-5 Pro could meaningfully speed up the path from question to insight.

The Takeaway

A capable model helped a respected immunology lab explain something real, and even anticipated experimental results it had never seen. That's a concrete, constructive example of AI-accelerated science working the way we'd hope: in service of human expertise, pointed at problems that matter for human health. It's exactly the kind of progress worth celebrating.

Sources: OpenAI — "GPT-5 Pro and an immunology mystery" — June 24, 2026; StartupHub.ai — "GPT-5 Pro solves an immunological puzzle" — June 24, 2026.