Barely a week goes by without a tech CEO making a bold claim about what artificial intelligence will do for medicine. Google’s president says we should be able to cure cancer in our lifetime.
The CEO of Arm Holdings says AI will crack it by modelling how cancer affects DNA at a scale no human could.
Sam Altman has suggested AI could compress decades of medical progress into a few years. The language is breathless, the timelines are short, and the headlines are irresistible.
Cancer specialists — the people who actually treat patients — are considerably less excited. And they have specific, well-grounded reasons why.
Cancer Is Not One Disease
The most fundamental problem with the promise of an AI cancer cure is the word “cancer” itself. Cancer is not a single disease. It is more than 200 biologically distinct diseases, each driven by different mutations, arising in different cell types, behaving differently in different patients, and responding differently to treatment. What cures one type may be entirely irrelevant to another.
Even within a single tumour in a single patient, the picture is more complicated still. Tumours are evolving ecosystems containing multiple genetically distinct clones. When treatment pressure is applied, it selects for resistant subclones — different regions of the same tumour can behave like different cancers. The idea that a single AI breakthrough could cut across this complexity and produce a universal cure fundamentally misunderstands the biology it claims to be solving.
What AI Can Actually Do
None of this means AI is irrelevant to cancer medicine — far from it. The genuine progress is real, and in some areas remarkable.
Google DeepMind’s AlphaFold has transformed structural biology, predicting the shapes of proteins with accuracy that previously required years of painstaking laboratory work. AI-assisted mammography has raised cancer detection rates meaningfully in large trials. AI tools are helping researchers identify promising drug targets, match patients to clinical trials more accurately, and predict tumour genetics from routine pathology slides that would previously have required additional expensive testing.
AI-assisted drug discovery has also cleared an important early hurdle: drug candidates developed with AI support are showing better safety profiles in early human trials than traditionally developed drugs — a genuine signal that the approach has value. One AI-designed drug, developed by Insilico Medicine for lung fibrosis, has reached Phase 3 trials, the most advanced stage before regulatory review. As of late 2026, no AI-designed drug has yet been approved, but the pipeline is building.
The Gap Between Discovery and Medicine
The problem is the distance between these genuine advances and the cure claims being made about them. AlphaFold predicts protein structures — it does not produce approved medicines. A drug candidate that looks promising in a laboratory must still pass through years of careful human trials, during which most candidates fail. Only around 4% of cancer drug candidates that reach human testing ever receive approval, after roughly 14 years of work and over a billion dollars in cost per drug. AI can speed up the early stages of that process. It cannot compress or eliminate the later ones.
The tumour microenvironment — the dense, hostile ecosystem surrounding cancer cells, complete with abnormal blood vessels, immunosuppressive cells, and biological conditions that block drug delivery — presents obstacles that better algorithms alone cannot overcome. Biology reroutes around blocked pathways. Even when AI identifies a promising target, cancers find alternatives. These are not data problems. They are biological problems that require new experiments, better drugs, and time.
The Real Risk of Overpromising
Cancer specialists aren’t dismissing AI out of hand. Most of them use it, and many are genuinely excited about specific applications. Their frustration is with timelines and framing that misrepresent what the technology can realistically do — and why that misrepresentation matters.
False hope is not a neutral mistake in cancer medicine. Patients and families make real decisions based on what they believe is coming. Research funding flows toward areas that appear close to a breakthrough and away from areas that appear stuck. Regulatory and healthcare systems face pressure to move faster than the evidence justifies. When the five-to-ten-year cure doesn’t arrive, the credibility of genuine progress gets caught in the backlash.
AI is changing cancer medicine. It is improving early detection, accelerating drug discovery, and helping match patients to treatments that are more likely to work for their specific biology. That is worth celebrating honestly. What it is not doing — and what no serious oncologist believes it will do on the timescales being discussed — is curing cancer.
The gap between “genuinely useful” and “cure for cancer in five years” is large. Closing it with hype doesn’t help the patients the technology is supposed to serve.