Your face is becoming a health score, and the rules have not caught up

AI can now read biological age and cancer survival from a photo, raising urgent questions of consent, bias and misuse long before any clinic adopts it.

Sign up for my Substack daily AI newsletter here.

Article link: https://johanosteyn.substack.com/p/your-face-is-becoming-a-health-score?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

For centuries, a doctor has glanced at a patient and read something in the face, a pallor, a tiredness, a frailty that the charts had not yet caught. That old clinical instinct is now being turned into a number. New research shows that an artificial intelligence model can estimate a person’s biological age from an ordinary photograph, and even predict how long a cancer patient is likely to survive. The science is striking, yet the harder question is not whether a machine can read our faces. It is what happens when it can, and who is allowed to use the result. Our faces are quietly becoming health data, and the rules that should govern that have not caught up.

CONTEXT AND BACKGROUND

The latest work comes from Mass General Brigham, published in a medical journal recently. Its system, called FAHR-Face, was trained on more than 40 million face images and drives two tools, one that estimates biological age and one that predicts cancer survival directly from a photograph. The researchers describe it as, to their knowledge, the first demonstration that an AI model can predict cancer survival from a simple picture, while stressing that the goal is to support personalised treatment decisions rather than to replace physician judgement.

This did not arrive from nowhere. It builds on a 2025 study in which cancer patients tended to look several years older than their actual age, and in which looking older was linked to worse survival. The new model is larger and was tested on an independent group of nearly 5,000 cancer patients, which is why it has drawn serious attention rather than dismissal.

INSIGHT AND ANALYSIS

The promise is real, and so are the warnings. Reviewing the study, Professor Miroslaw Bober of the University of Surrey called it a convincing statistical link but a proof of concept that is “not ready for clinical use”, noting that predicting an outcome is not the same as explaining it, and that smiling or a change in lighting shifted the AI’s age estimate by about two years. A tool that moves two years because someone smiled is not yet a verdict about a human life.

One finding should concern South African readers most. In the same review, the facial-age measure was not significantly associated with survival for Black patients. A model trained largely on faces unlike those of most South Africans may simply read our population less well, and importing it into African clinics without local validation would risk building that bias into decisions that are a matter of life and death.

There is also the question of what a face reveals and who may look. I have previously written about this, arguing that South Africa’s largest facial recognition networks already run through private hands with little real redress, and that the Information Regulator now treats facial recognition as prohibited by default under our data protection law. A face is biometric data, among the most sensitive categories the law recognises. Once a photograph can estimate how fast you are ageing, an insurer, an employer or a lender would dearly like to see it, and our rules were not written for that.

IMPLICATIONS

For boards, regulators and clinicians, the lesson is to govern the signal before the shortcut becomes routine. A facial health score is evidence for a trained professional to weigh, never an automated judgement about a person. Any clinical use should be validated on local and diverse faces first, and tested against doctors’ own assessments, not slipped in to plug a gap. Regulators should treat these scores as the sensitive biometric data they are, with clear consent, purpose limits and a firm line against their use in cover, credit or hiring.

The temptation here is specific to our context. South Africa’s cancer services are stretched and deeply human, with palliative teams at hospitals such as Groote Schuur working to protect dignity when cure is no longer possible. In a system under that much pressure, a cheap tool that triages from a selfie will look like relief. It can help only if it supports the clinician and the patient, and never quietly replaces the judgement and the care that give medicine its meaning.

CLOSING TAKEAWAY

The face is the most personal biometric we carry, holding our ageing, our health and now, apparently, our odds. The science is moving faster than the safeguards, and that gap is where the real harm will happen, in a denied policy, a quiet rejection, a number trusted more than a doctor. We should welcome tools that help clinicians see more clearly, while insisting that a photograph is a clue and never a sentence. What must stay in human hands is the judgement about what any of it means, and the dignity owed to the person in the picture.

Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net