OpenAI just helped solve 18 medical mysteries. These were cases doctors had been stuck on for years.
Every one of them was a rare genetic disease in a child. Families who had spent years, in some cases decades, with no name for what was wrong. And in a single study, that changed.
This is one of the most hopeful things I have read in medicine all year. Let me break down what actually happened.
Until now, diagnosing a rare disease came down to one thing: a doctor's memory
When a child has a rare genetic condition, the process has always been the same. An expert reads their genes, studies their symptoms, and tries to match them to a disease they already know.
But here is the problem. There are more than 7,000 known rare diseases, with more found every year. Underneath them sit millions of tiny genetic variations. No single doctor, however brilliant, can hold all of that in their head.
So the answer is often already there in the data. There is just too much of it for any one human to connect. And that is why roughly half of these children never get a diagnosis, and their families are left wondering what is wrong.
Then OpenAI's o3 model did what no human could
Researchers at Boston Children's Hospital, Harvard and OpenAI took 376 old cases that had gone unsolved. Each had already been through genetic testing and expert review. For all practical purposes, these files were closed.
They handed them to OpenAI's o3 reasoning model. It read each child's genes, their symptoms, their family inheritance pattern, and every relevant study ever published, all at the same time.
Then it connected clues and patterns across all of it that no human doctor had ever thought to try. In one case, it even spotted a problem in a child's DNA sequencing that was not labelled anywhere in the data.
It did not just match things together. It reasoned.
Doctors checked every lead, and confirmed 18 brand new diagnoses
Here is the part that makes it real. The doctors stayed in charge the whole way through. They reviewed every single lead the model surfaced, ran the tests, and confirmed 18 brand new diagnoses.
That is almost 5% of cases everyone had already given up on. As one of the lead researchers put it, that might not sound like a lot, until you remember how many times these files had already been analysed by experts. Each one is an answer for a family that had waited years to hear it.
That is the headline here. AI and doctors, working together, cracked cases that neither could have cracked alone.
For one family, this was the call they had waited fifteen years for
It is easy to talk about 18 and forget what each one means.
One patient started showing symptoms around age nine. Slowing down in sport. Walking up on her toes. By thirteen she was in a wheelchair and on a ventilator. Specialists across multiple institutions looked, and found nothing. Then the AI revisited genomic data everyone had already set aside, and surfaced the single variant that finally gave her condition a name.
The call to her family started with a line you do not forget. We know it has been about fifteen years, but we have some news.
I trained as a doctor. I know the weight of telling a family you do not have an answer. This is the opposite of that, and it is the whole reason this work matters.
This proves AI can become a memory of record for medicine
Here is why this is a genuine breakthrough.
For the first time, AI has shown it can team up with doctors to become a memory of record, holding the answers that no single human ever could. It does not get tired. It does not forget. It can hold thousands of diseases, every variant, and every new study at once, and bring the right one to the right doctor at the right moment.
That means faster diagnoses. New disease mechanisms we have never spotted before. And new hope for millions of families still searching for what is wrong with their child.
And this is only the beginning
It feels small right now. Eighteen children. One study. But this is the start of something far bigger.
When AI can hold the entire record of medical knowledge and reason across it alongside a doctor, every undiagnosed case becomes a case that might be solvable tomorrow. This is the beginning of a wave of scientific breakthroughs, and we are watching the very first one.
The answer was always there. We finally have a way to find it.
So here is my question. If AI can now hold the entire record of medical knowledge, which disease do you most want it to crack next? Tell me below, and I read every comment.