Sick of hearing about AI? You’re not alone. There are, after all, only so many hours we can spend scrolling through AI slop on our phones before wondering if this potentially species-propelling tech could be put to better use. For example – using AI to help scientists find better cancer treatments. Not a bad place to start, eh?
Thankfully, there’s already some encouraging work happening. In a large Swedish trial, for example, AI helped doctors find more breast cancers while also reducing the number of scan readings required. Win-win. Elsewhere, researchers are using it to investigate new treatments, with some predicting that decades of medical progress could happen within a few years.
It’s certainly an appealing thought, and a far better use case for the tech than the aforementioned slop. Having said that, there’s plenty of work yet to be done. So with that said, how much closer is AI really getting us to curing cancer? Here are but a few examples.
AI is already helping doctors find cancers
At the risk of repeating the header, AI is already doing great work. In Sweden, mammograms are normally read by two radiologists, but for the MASAI trial, researchers used AI to help decide which scans needed both readers, and which could go to one. The software also marked suspicious areas for the doctors to examine.
In total, more than 105,000 women were randomly assigned to the AI-supported approach or standard screening. Results published in 2025 showed that AI-supported screening detected 29 per cent more cancers – roughly 6.4 cancers per 1000 women, compared with five per 1000 in the standard group.
That also didn’t come with a statistically significant increase in false positives – cases where someone is called back for further checks but doesn’t have cancer. And because some scans only needed one reader, the number of readings fell by around 44 per cent.
Follow-up results published have since found that AI-supported screening detected around 81 per cent of the cancers diagnosed during the study period, compared with 74 per cent for standard screening. Whether that translates into longer lives though, remains to be seen.
It’s still early days
Results like the MASAI trial help explain the interest in medical AI, while some developers expect progress on a much greater scale.
Enter Dario Amodei, the boss of Anthropic – the company behind Claude. He outlined one such prediction in his 2024 essay Machines of Loving Grace. He suggested that powerful AI could squeeze 50–100 years of progress in biology and medicine into five to ten years. Eliminating most cancer was among the possibilities he described, although he acknowledged that some difficult cancers could remain.
There’s a fairly big condition attached to that prediction, mind. His five-to-ten-year window only begins once a much more capable AI system than today’s familiar chatbots has arrived. That technology has to be developed first – so the prediction tells us little about when a particular treatment might reach patients.
It’s also worth remembering that many cancers can already be cured when detected early and treated effectively. The challenge is making treatment successful for more people, across a huge range of diseases.
The US National Cancer Institute also describes more than 100 types of cancer, and each person’s cancer has its own combination of genetic changes. Even cells within the same tumour can differ. A treatment that works for one might not work for someone else, which is why it’s important to keep a level head and view this all as a marathon, not a sprint.
Beyond diagnosis, AI could also help researchers develop new drugs. That can involve finding a biological target – such as a protein involved in a disease – and designing a molecule to act on it.
Searching for promising molecules can, unsurprisingly, take a considerable amount of time. According to researchers at the US National Institutes of Health, screening a large collection of compounds can take months or years, while computer-based screening can sometimes do the job in days or even hours. That gives scientists a shortlist to investigate in the lab much sooner – although they still have to establish whether those molecules work in practice.
For example, a collaboration between the US National Cancer Institute and Department of Energy used AI with molecular simulations to study RAS proteins, which are frequently altered in cancer. The aim was to understand how they interact with other proteins and the cell membrane, helping researchers work out how to target them with treatments.
Once a potential cancer drug reaches human trials, early studies investigate safety, dosage, and whether it appears to work. Larger phase-three trials compare it with standard treatment – allowing researchers to assess whether it offers patients something better, and whether its benefits justify its side effects.
Even if AI helps scientists reach those trials sooner though, patients still need to be followed over time. Researchers have to find out how long the benefits last and what problems might emerge later.
Data matters too
The information used to train an AI also affects how useful it will be. The National Cancer Institute warns that models can perpetuate medical bias when their training data doesn’t properly represent the wider population. People who were poorly represented in that data also still need the resulting tool to work for them.
Testing also matters for the software, as well as the drugs. A good result on one set of scans or patient records doesn’t establish that the same tool will perform equally well elsewhere. Researchers need reliable data and clinical studies to find out where it helps, where it makes mistakes, and how it should fit into care.
Ultimately, there’s no shortage of evidence that gives us reasons to take AI and its role in curing cancer seriously, but we’re a long way away from making one of humanity’s biggest ever breakthroughs. For now, it’s best to keep a level head, and follow the science. As the tech evolves, so too will our methods and discoveries. And if AI lives up to the hype, we’re right around the corner from a whole bunch more…
