Shyam Sankar, chief technology officer of the U.S. data analytics company Palantir, has drawn attention by saying that some AI researchers have taken up smoking to show their confidence that AI will advance cancer treatment. He made the remark on CBS's "Face the Nation," broadcast on October 11, and it can also be confirmed in the interview transcript published by CBS. However, no survey or specific example was offered to show how many such researchers actually exist.
This odd anecdote highlights a larger question: how far can expectations for AI-driven medical progress justify decisions made today? Research that uses AI to predict cancer patient survival and clinical trials of AI-designed drugs are indeed under way. But what each study has actually demonstrated differs, and none of it guarantees future treatment effects. Even if AI transforms medicine, the time needed to confirm a treatment's efficacy and safety in the human body cannot be ignored.
Are there really researchers who smoke because "AI will cure cancer"?
Sankar described a story in which some AI researchers start smoking to show strong faith in future medical technology. He cited it as one example of an excessive devotion to AI.
Still, this remark alone does not confirm that any researchers have actually behaved this way. No specific individuals or survey results were presented, and there is no basis for concluding that smoking has become widespread among AI researchers.
Even if some people did start smoking to show confidence in their own research, the behavior would prove only that they accepted the risk themselves. It would not support the probability that AI will one day cure cancer, or that the necessary treatment will arrive within their lifetimes.
That said, predictions that AI will dramatically advance cancer treatment have been made publicly by prominent AI researchers and business leaders. A closer look at those predictions shows that even this optimism rests on important preconditions.
Anthropic's CEO predicts cancer deaths and incidence could fall by more than 95%
In his October 2024 essay "Machines of Loving Grace," Dario Amodei, CEO of Anthropic, argued that powerful AI could compress 50 to 100 years of progress in biology and medicine into 5 to 10 years.
In Amodei's essay, he also predicts that cancer incidence and mortality could eventually be reduced by more than 95%. At the same time, he acknowledges that complete conquest of cancer would be difficult, because there are so many types of cancer and because it adapts readily, for example by acquiring resistance to treatment through mutation. He notes that some rare and hard-to-treat cancers may remain.
The key point is that Amodei's prediction is not premised on today's AI technology.
What he envisions is a world after the arrival of extremely advanced AI, with intellectual capabilities surpassing Nobel laureates across a wide range of fields and able to direct complex experiments and research. In other words, the "5 to 10 years" is not a deadline counted from 2024; it refers to the time after such powerful AI exists.
The "95% or more" figure is also not the result of a clinical trial in which a particular drug cured 95% of patients. It is a forecast of how far society-wide incidence and mortality could fall if cancer prevention and treatment advance dramatically.
Amodei himself also points out that, no matter how far AI's intelligence improves, some experimental time cannot be shortened.
For example, experiments with cells and animals, clinical trials to confirm a drug's effects in the human body, and studies of long-term health effects all require a certain amount of time. Running many studies in parallel with AI could speed up science as a whole. But the sequence of checking one experiment's results before designing the next cannot be eliminated entirely.
In short, even those with strong hopes for an AI-driven medical revolution acknowledge that it takes time to get from research to actual treatment.
Expecting cancer treatment to advance greatly in the future is clearly different from assuming that you will receive the treatment you need when you fall ill.
AI is already producing results in cancer research, but "prediction" and "treatment" are different things
AI's contribution to medical research is not merely a forecast. Concrete findings have already been reported in areas such as predicting cancer patient survival and developing new drugs.
One example is the AI model "scSurvival," developed by Tao Ren and colleagues. According to the research paper published in the peer-reviewed journal Cancer Discovery, the model analyzes gene expression information from individual cells that make up a tumor together with patient survival data.
According to an explainer published by the U.S. National Cancer Institute (NCI) on April 21, 2026, the research team evaluated the model using clinical data from more than 150 cancer patients. It analyzed data on malignant melanoma and liver cancer, predicted patient survival, and identified cell populations associated with survival rates.
A distinguishing feature of scSurvival is that, rather than simply averaging information across a whole tumor, it considers how strongly each individual cell relates to patient survival. By weighting cells that are strongly linked to survival, it can capture differences among cell populations that are hard to see in tumor-wide averages.
Such technology could improve risk assessment for individual patients and provide more information for considering treatment plans. However, what this study confirmed was survival-prediction performance; it did not show that AI-based treatment extended patients' lives.
There are also cases in which AI has been used to develop a new drug that has advanced to clinical trials in patients.
One is rentosertib, a drug candidate from Insilico Medicine. In the Phase 2a clinical trial paper published in Nature Medicine in June 2025, the drug's safety and efficacy were evaluated over 12 weeks in 71 patients.
However, rentosertib's target disease is not cancer but idiopathic pulmonary fibrosis (IPF), in which lung tissue gradually scars and breathing function declines. The trial was conducted as a placebo-controlled study, with patients randomly assigned to groups and using a double-blind design in which neither patients nor researchers knew who received what.
The differences between the two studies can be summarized as follows.
| Study | Subjects and method | What was actually evaluated | What has not been demonstrated |
|---|---|---|---|
| scSurvival (2026, Ren et al.) | Analyzed single-cell data from malignant melanoma and liver cancer together with patient survival information | Survival-prediction performance and cell populations related to survival | That AI-based treatment extends patients' survival |
| rentosertib (2025, Xu et al.) | 12-week placebo-controlled trial in 71 patients with idiopathic pulmonary fibrosis | Primarily adverse events, with changes in lung function measured as a secondary outcome | Efficacy against cancer, long-term life extension, or cure of disease |
scSurvival is research that aims to predict patient prognosis more accurately with AI. Rentosertib, by contrast, is research in which a drug candidate developed with AI was given to actual patients and evaluated. Both are achievements of AI use in medicine, but because they test different things, their success rates cannot be compared by the same standard.
Signs of improved lung function with an AI-designed drug
In the rentosertib clinical trial, signs of improvement in lung function were observed.
Forced vital capacity (FVC), one measure of lung function, measures how much air a person can forcefully exhale after inhaling as deeply as possible. In the trial, the group of patients given 60 mg of rentosertib once daily showed an average increase of 98.4 mL from baseline. The placebo group, meanwhile, showed an average decrease of 20.3 mL.
The 95% confidence intervals were 10.9 to 185.9 mL for the 60 mg rentosertib group and −116.1 to 75.6 mL for the placebo group.
However, the trial's primary endpoint was the occurrence of adverse events, and the change in lung function was a secondary endpoint. In addition, 16 of the 71 participants did not complete the planned treatment, and treatment discontinuations related to liver toxicity and diarrhea were also reported.
The results show that a drug developed with AI has advanced to the clinical trial stage, where its effects on patients can be evaluated. But changes in lung function over 12 weeks cannot be said to confirm long-term life extension or a cure. Still less is it evidence that the drug can treat cancer.
Progress in AI-driven drug discovery and whether its results can already be used to treat a wide range of diseases must be evaluated separately.
Even if AI conquers cancer, the harms of smoking will not disappear
Even if AI-driven cancer treatment advances dramatically in the future, that is no reason to downplay the risks of smoking.
To begin with, cancer is not a single disease. The NCI explains that there are more than 100 types of cancer, each with different characteristics depending on the organ, tissue, and cell type in which it arises.
Even if an effective treatment is developed for one type of cancer, it will not necessarily work for all others. There is also no guarantee that the various cancers that smoking can cause will someday be overcome by a single therapy.
In its explanation of smoking and cancer, the U.S. Centers for Disease Control and Prevention (CDC) says smoking is linked to about one-third of cancer deaths in the United States.
This does not mean that one-third of smokers die of cancer. It is an estimate of how much smoking contributes to cancer deaths across society as a whole.
Harmful substances in tobacco smoke damage cellular DNA and also impair immune function. The effects are not limited to the lungs and can lead to cancers in many organs. Continuing to smoke after a cancer diagnosis can also raise the risk of death, reduce the effectiveness of treatment, and increase complications.
More important still, the health harms caused by smoking are not limited to cancer.
According to the CDC's explanation of cardiovascular disease, smokers' risk of coronary artery disease and stroke is estimated to be about two to four times that of nonsmokers. Causes include damage to blood vessels from smoking and a greater tendency of blood to clot.
Chronic obstructive pulmonary disease (COPD) is another serious harm of smoking. In COPD, breathing function declines as the walls of the alveoli are destroyed and the airways become inflamed. In its explanation of smoking and COPD, the CDC says that continued smoking tends to accelerate lung damage, while quitting can slow disease progression and the decline in lung function.
Even if AI someday makes it possible to treat many cancers, that alone cannot be said to restore blood vessels and lung tissue damaged by smoking.
Furthermore, the harms of tobacco smoke are not confined to the smoker. Secondhand smoke creates risks of lung cancer and cardiovascular disease for nonsmokers nearby.
Smoking in the belief that future medical technology will save you means not only accepting health risks yourself but also potentially affecting the health of the people around you.
Even if AI speeds up drug development, clinical trials take time
In considering AI-driven medical progress, the time between the discovery of a new treatment and the point at which patients can actually use it also matters.
In its explanation of clinical trials, the U.S. Food and Drug Administration (FDA) says that in typical drug development, Phase 2 trials take several months to two years and Phase 3 trials take roughly one to four years.
Phase 3 trials are generally the stage at which a treatment's efficacy and safety are confirmed in hundreds to thousands of patients. According to the FDA, participant numbers typically range from about 300 to 3,000.
Of course, the required duration and number of participants vary with the type of drug and target disease. But large, long trials play an important role in examining side effects that are hard to detect in early stages and how long a treatment's effects last.
Even if AI can quickly identify promising treatment candidates, that does not remove the need to verify whether a candidate is safe for humans and can actually improve disease.
A former cancer researcher, interviewed anonymously by the British IT news site The Register, also took a cautious view on this point.
According to The Register's article dated October 14, 2026, the former researcher pointed out that even with AI, it could take several years for new cancer drugs to be used in actual medical practice. The former researcher also explained that, after a drug is developed, whether tumors acquire resistance to it must be checked separately.
In other words, the speed at which AI discovers promising drugs differs from the time it takes to confirm that those drugs can save lives over the long term.
To accept present health risks while counting on future treatments, it is not enough for AI simply to discover excellent therapies. By the time you need the treatment, its safety and efficacy must have been confirmed and it must be available in real-world medical care. It must also be suited to your particular disease.
Various studies have already begun to show that AI could greatly advance medicine. But the expectation that science as a whole will progress cannot be turned into a guarantee that any individual will receive appropriate treatment in the future.
In evaluating the results of AI medicine going forward, it will be important to look at which patients were studied, what was tested, and how large a treatment effect was confirmed over what period.
If AI-based survival prediction helps in choosing appropriate treatment, and drugs developed with AI show lasting effects in large clinical trials, that progress could lead to medical technologies that actually save patients' lives.