Can AI cure cancer?
AI has already changed the chemistry, and it has a Nobel Prize to show for it. What it has not yet done is get a drug it discovered approved.
Key figures
The 2024 Nobel Prize in Chemistry went to protein structure prediction and computational protein design.
One system predicted structures for around 200 million proteins, effectively the known protein universe.
AI-discovered molecules succeed in Phase I trials at 80 to 90%, then fall to about 40% in Phase II, which is the historic industry average.
What AI has already won
This is not a field where the achievements are speculative. The 2024 Nobel Prize in Chemistry was awarded half to David Baker for computational protein design, and half jointly to Demis Hassabis and John Jumper for protein structure prediction.
The scale of the second is worth stating plainly. Having established that the system worked, its authors predicted the structures of essentially every protein science had catalogued — around 200 million of them. A problem that took a doctorate per protein became a lookup.
That is a real and enormous change to how drug discovery starts. It is also not the same thing as a cure.
Where it stops
Molecules discovered or designed with AI have been succeeding in Phase I trials at roughly 80 to 90%, well above historic industry averages. Phase I asks whether a compound is safe and behaves in a human body as expected, and AI is evidently good at designing molecules that do.
In Phase II the rate is about 40%, on a limited sample, which is comparable to the historic industry average. Phase II asks the different question: does it actually work on the disease.
As of July 2026, no drug discovered or designed by AI had received full approval from the United States Food and Drug Administration.
The translational gap
The distance between those two bars has a name. The translational gap is the space between predicting a molecule's properties and predicting whether it helps a sick person.
AI has made enormous progress on the first, because the first is a structured prediction problem with abundant training data. The second is a question about a human being with a disease that is itself heterogeneous, and there is no dataset of counterfactual patients to learn from.
This is why both popular framings are wrong. “AI will cure cancer” ignores that the hard part was never the chemistry. “AI drugs keep failing” ignores that they fail at the same rate everything else fails, having got there faster and cheaper.
So can it cure cancer
Not on its own, and not as a single event. Cancer is not one disease, and nothing about the current evidence suggests a system will output a cure.
What the evidence does support is narrower and still significant: more candidate molecules, designed faster, reaching trials in better shape, against targets that were previously inaccessible because nobody knew the shape of the protein. That shifts the odds at the front of a pipeline whose back end is unchanged and still takes years.
The honest one-line answer is that AI has already made real cancer drugs more likely, and has not yet produced one.
Sources and method
- The Nobel Prize in Chemistry 2024, awarded to David Baker for computational protein design and jointly to Demis Hassabis and John Jumper for protein structure prediction. Source of the 2024 prize and its division.
- AlphaFold Protein Structure Database, EMBL-EBI and Google DeepMind, the public database holding the predicted structures referred to here.
- How successful are AI-discovered drugs in clinical trials? A first analysis and emerging lessons, Drug Discovery Today. Source of the Phase I and Phase II success rates for AI-discovered molecules, on a limited sample.
- The statement that no AI-discovered drug had full FDA approval is as of July 2026.
- Every source here was opened and checked on 25 September 2026. Where nothing has been measured, this page uses a table and says so rather than drawing a chart of an opinion.