We have always understood illness through its arrival: a persistent cough, an unfamiliar pain, the unsettling discovery of an abnormal blood test. But what happens when illness arrives before the disease itself? When your body feels perfectly healthy, yet artificial intelligence has already written a future in which it is not? What cruelty, to be made to answer for a disease you may never develop.
Imagine receiving two notifications on an otherwise ordinary morning.
The first comes from your healthcare provider. An artificial intelligence (AI) system has identified patterns suggesting an increased likelihood of developing a serious illness. There are no symptoms, not even a diagnosis, just a probability. The second comes from your insurer. Because of the AI’s prediction of what could happen, your premiums have increased and your coverage narrowed. You have not fallen ill, and perhaps never will. Yet somehow, you have already become a financial liability, to be calculated as a line-item loss on your insurance company’s balance sheet.
This hypothetical exposes an uncomfortable possibility in AI-driven medicine. The technology that promises to protect us from tomorrow’s illnesses could allow institutions to penalise us for them today.
In September 2025, European researchers introduced Delphi-2M, an AI model capable of estimating risks across more than 1,000 diseases and simulating possible health trajectories extending two decades into the future. Trained using approximately 400,000 UK Biobank participants and validated against Danish health records, it represents an extraordinary scientific development: teaching machines to recognise patterns in illnesses that have not yet occurred. However, prediction is not prophecy, regardless of how intelligent machines learn. These estimates describe probabilities, not predetermined destinies. Their reliability varies, and findings from European populations cannot simply be transplanted into Malaysia’s ethnically diverse society. Nevertheless, a machine need not predict your future perfectly to influence how others treat you in the present.
Professor Dr Thong Meow Keong, a Malaysian clinical geneticist, cautions that AI depends on the data from which it learns. “AI is basically a tool,” he said in our interview on 20 October 2026. Drawing on genomic medicine, he described the difficulties of applying findings learned from predominantly Western populations to Asian and African populations. “They come up with a lot of uncertainties,” he said. His example offers a parallel for AI: a model trained on one population cannot simply be assumed to perform well in another. The World Health Organization has likewise cautioned that systems trained mainly on data from high-income countries may perform poorly in other settings. Models intended for Malaysians need to be tested using data that reflects the people and healthcare settings they will serve.
Of course, medicine has long attempted to anticipate illness. Genetic testing can reveal inherited vulnerabilities decades before symptoms emerge, while conventional risk calculators estimate future disease using factors such as age, cholesterol and blood pressure. Neither medical prediction nor discrimination based on anticipated illness began with AI. What AI introduces is a potentially unprecedented scale and scope of prediction. Rather than assessing particular inherited conditions or familiar risk factors, systems such as Delphi-2M can examine sequences of medical diagnoses and estimate risks across hundreds of diseases simultaneously. Future applications could incorporate wearable devices which routinely collect health data, thus continually revising predictions without requiring genetic testing. Such assessments could be automated across enormous populations, potentially at relatively low additional cost.
What was once an occasional assessment could become an evolving portrait of someone’s future illnesses, constructed long before they experience a single symptom.
Here lies AI medicine’s emerging dilemma. For a clinician, elevated risk may justify earlier screening, closer monitoring or preventive treatment. For an insurer, the same information could justify higher premiums, restricting coverage or triggering reconsideration of whether someone is worth insuring. The prediction remains unchanged. The only difference is the interests of those interpreting it. The World Health Organization has already warned that insights derived from AI and machine learning could be misused to exclude expensive patients or conditions from healthcare coverage, or to increase insurance premiums. Technology intended to improve healthcare could instead deepen existing inequalities. But perhaps the most disturbing possibility is not that AI might predict incorrectly. It is that AI might predict correctly.
Much of our anxiety surrounding medical AI concerns its failures: missed cancers, discriminatory algorithms and fabricated hallucinations masquerading as medical advice. Yet even a flawlessly accurate prediction cannot answer whether someone should bear financial consequences for an illness they have not developed, particularly one they may have little power to prevent. Scientific accuracy does not establish moral legitimacy. A probability is not a diagnosis, nor should it automatically become a verdict on someone’s entitlement to protection.
Insurance, admittedly, has always involved assessing risk. Insurers must remain financially sustainable, and sophisticated forecasting could improve resource allocation and preventive care. However, insurance also depends upon the sharing of uncertainty. We contribute while healthy because nobody knows who will eventually require help. What happens to that collective promise when AI allows institutions to cheaply distinguish between those likely to remain healthy and those who may become expensive? Hence, a population might become easier to predict without becoming easier to protect.
For Malaysia, this question arrives at a particularly vulnerable moment. In December 2024, Bank Negara Malaysia cited medical cost inflation of 15%, prompting interim measures to ease rising medical insurance and takaful premiums. Healthcare affordability was already under pressure, long before these predictive technologies became a realistic consideration. Now picture a young Malaysian undergoing AI-assisted health screening. The system identifies an increased future risk of diabetes or cardiovascular disease. Clinically, this might be invaluable information. Financially, should such predictions influence insurance underwriting, the same discovery could become a lasting disadvantage.
Malaysia is also investing in data that better reflects its own population. The government’s MyGenom project, for example, aims to build a genomic reference dataset reflecting the country’s diversity. It is one example of efforts to strengthen local evidence, although genomic data differs from the health records used by models such as Delphi-2M. Nevertheless, the Ministry of Digital, primarily through the newly established National AI Office (NAIO) is actively accelerating the adoption of AI-driven diagnostic tools across the national healthcare landscape. These policy pushes suggest that predictive medicine is no longer a distant future, but an imminent reality coming to our population. However, as the possibility of integration between MyGenom’s dataset and local electronic health records draws closer, it also heightens the risk that without strict ethical or legislative guardrails, this same union of data could easily be weaponised to pre-emptively price citizens out of healthcare based on their digital genetic and healthcare profiles.
Looking ahead to the completion of MyGenom, Professor Thong raised a concern about future access to the data. “We hope that they won’t sell it to the industry,” he said. In the context of the interview, his concern included insurers gaining access to information collected to improve health. MyGenom’s published guidance describes approval requirements for research access and safeguards against discriminatory use. For AI-based healthcare more broadly, the question is what institutions should be permitted to do with increasingly detailed information about future health. Better Malaysian data may improve predictions, but it does not settle whether those predictions should determine access to financial protection.
An even crueller paradox surfaces if we dwell deeper. What if people begin fearing the consequences of knowing their own health risks? What if someone declines preventive screening, avoids sharing medical information or hesitates before consenting to an AI assessment, not because they distrust medicine, but because they distrust what might happen to its findings? We would have created a healthcare system in which ignorance offers a kind of financial protection. The consequences could extend beyond individual policies. If predictive AI allows private insurers to select increasingly favourable risks, patients with more complex anticipated needs could become more dependent on Malaysia’s already strained public healthcare system. The financial burden would not disappear, it merely changed hands into the taxpayer’s purse.
Malaysia must therefore decide what purpose these technologies should serve. Medical information collected to improve care should not become an unrestricted resource for unrelated financial decisions. Regulators must scrutinise not merely what information insurers collect, but what AI and machine learning algorithms are permitted to infer from it. After all, discrimination based on vulnerability may no longer require a genetic test. Consequential algorithmic assessments must remain contestable and subject to human oversight. Regulators must distinguish between using AI to reduce someone’s likelihood of illness and using it to reduce an institution’s responsibility towards that person.
There is another future worth pursuing. Insurers and takaful operators could use predictive AI to support earlier interventions, subsidise preventive screening and help policyholders remain healthier. Demonstrating that better predictions could become instruments of solidarity rather than exclusion. Medicine’s purpose has never been simply to recognise disease. It is to respond to human vulnerability with knowledge, compassion and responsibility.
And so we return to that ordinary morning. Two notifications illuminate the same possible future. One offers an opportunity to intervene before illness arrives. The other threatens to make that future more difficult to survive. Neither can tell you whether you will actually become ill. But together, they reveal something far more certain about the society receiving them. Perhaps the greatest test of artificial intelligence in medicine will not be whether machines can predict our suffering decades before it begins. It will be whether, having foreseen that suffering, we choose to protect the person who might endure it.
Otherwise, AI gives you an early warning. Your insurer sends you an early bill. Welcome to preventive healthcare.
AI could soon predict illnesses years before symptoms emerge. But if insurers gain access to these predictions, could early warnings become grounds for higher premiums or restricted coverage?
About the Author: Yap Khai Chuen is a medical student and a fellow at The Petri Dish Writing Lab. Passionate about medical ethics and evidence-based medicine, his writing addresses public health issues, emerging healthcare technologies, and the challenge of medical misinformation.
