Your AI Agent Is About to Know More About You Than Your Doctor Does

It is a Tuesday evening in Dubai. For three weeks your resting heart rate has crept upwards by a few beats. Your calendar shows back-to-back flights to Riyadh and London, your food deliveries have drifted towards late-night orders and a borderline vitamin D result from your last blood panel is still waiting for a follow-up. Each signal lives in a different app, beyond your doctor’s view. A personal health agent would see all of them – and might suggest a check-up before you notice anything yourself.

The first health agents are already taking shape

On 20 September, Axios described the arrival of a new generation of personal AI agents that manage calendars, make calls, shop, book tickets and navigate online services on their owner’s behalf. Their defining advantage is memory. The more an agent knows about someone’s schedule, relationships, preferences and spending, the more effectively it can act for them.

A chatbot answers questions. An agent keeps context, notices change and moves a process forward – reminding, comparing, booking or escalating within the authority it has been given.

Preventive-health companies are already building on this model. Australian healthtech company Everlab recently introduced Concierge, centred on an AI health assistant called Eva. According to the company, Eva works with a member’s health profile, results, medical history, wearable data and preferences. A member can ask how travel might affect a training programme, whether a supplement fits an existing plan or when a biomarker should be retested. When clinical care is required, the service connects them with Everlab’s doctors, nutrition specialists and exercise professionals. The agent holds the context; the clinicians hold responsibility for care.

Everlab’s figures show why longitudinal data has become so valuable. The company says it has completed more than 40,000 consultations for over 20,000 patients and processed more than 21 million biomarker results. In June it announced an AU$65 million Series A round to fund international expansion and its preventive-care infrastructure.

According to OpenAI, more than 300 million people already bring health questions to ChatGPT every week

The largest consumer-AI companies are moving in the same direction. Health in ChatGPT , which rolled out to US users in July 2026, can connect Apple Health and supported medical records. With permission, it places medications, laboratory results, recent appointments, sleep and activity inside the context of a conversation. According to OpenAI, more than 300 million people already bring health questions to ChatGPT every week.

Wearables are becoming a continuous data layer. AI now supplies the interpretation, the memory and the conversation.

Breadth versus judgement

The agent’s strength is breadth and continuity. It can process years of records, thousands of wearable measurements and patterns too gradual for any person to remember. A clinician brings a different kind of intelligence: medical training, physical examination, a feel for ambiguity, experience with atypical presentations and responsibility for the consequences of a decision.

More data can also breed false confidence. Wearables produce gaps and artefacts. Medical records carry old diagnoses and medications a patient stopped taking long ago. A rise in heart rate may reflect illness, anxiety, heat, travel, alcohol, a faulty sensor or several of these at once. An agent can spot a correlation and misread its cause.

The companies building these products acknowledge the boundary. OpenAI states that its health features can make mistakes and are designed to support professional care. The most credible model is a collaboration: the agent maintains the longitudinal picture, the patient supplies lived context and the clinician carries medical judgement and accountability.

The Gulf’s head start – and its test

The Gulf offers a revealing case. The UAE has spent years building connected health records: Dubai’s NABIDH and Abu Dhabi’s Malaffi platforms link hospitals and clinics into shared patient histories, and the Emirati Genome Programme has added genomic data at national scale. The UAE’s law on the use of ICT in health fields (Federal Law No. 2 of 2019) requires health data to be stored inside the country, with exceptions set by regulators. In Saudi Arabia, the Personal Data Protection Law, fully enforced since September 2024, classifies health data as sensitive.

For a global agent, these rules raise practical questions. Where does a Dubai resident’s health memory live? Which jurisdiction governs it? And what happens when an agent built in California reasons over data that UAE law expects to remain in the UAE? The region’s investment in longevity and preventive medicine makes the answers especially consequential here.

Elsewhere, the picture is more fragmented. In the United States, the Federal Trade Commission explains that HIPAA generally applies to covered healthcare organisations and their business associates. Health information held by an independent consumer app may fall outside it, even when that information originally came from a provider. Other consumer-protection and state privacy laws may still apply, yet the common assumption that all health information enjoys equal legal protection is mistaken.

Europe is taking a more structured route through the European Health Data Space , which gives individuals greater control over access to their electronic health records and restricts certain secondary uses, including marketing and decisions that could harm an individual. Major provisions begin to apply from 2029.

Personal agents complicate every one of these frameworks because they dissolve the border between medical and non-medical data. The system stores a cholesterol result and then connects it with food orders, exercise, sleep, work pressure and medication adherence. The health profile lives in the relationship between those sources.

Consent has to become granular

Today, digital consent usually means accepting a privacy policy few people read. That model breaks down once an agent can continuously combine data and initiate actions.

A person may be comfortable letting an AI analyse sleep and exercise while keeping reproductive-health records private. They may want the agent to suggest an appointment and keep the decision to book it for themselves. They may allow a summary to reach their doctor and bar access by an employer, insurer or advertising platform.

Consent therefore needs to be specific, visible and reversible. Users should be able to see which sources an agent has accessed, what it inferred, which actions it took and who received the information. Permissions should work in layers, with the rights to read, interpret, recommend, share and act each requiring a different level of authority.

The right to export, correct and delete an agent’s memory will matter as much as the right to access a medical record

Health-agent memory also raises a question conventional medical records rarely face: what happens when the user changes platforms? After years of learning a person’s history, preferences and behavioural patterns, the agent’s accumulated context may be worth more than the original data. Losing it could create a powerful form of dependency.

The right to export, correct and delete an agent’s memory will matter as much as the right to access a medical record. Users will also need to tell facts from inferences. “The patient takes this medication” is a record. “The patient may be entering a depressive episode” is an interpretation with very different consequences.

The doctor will meet the version of you the agent has assembled

The personal health agent will change the medical consultation well before it reshapes clinical practice itself.

Instead of reconstructing six months from memory, a patient could arrive with a structured account of what changed, when it changed and which interventions seemed to help. The doctor could spend less time gathering scattered information and more time interpreting it. Preventive care could become continuous, and subtle changes could surface before they turn into obvious symptoms.

To provide that continuity, the agent must build an unusually complete model of a human life. It may know the body through laboratory tests, the mind through conversations, behaviour through purchases and movement through devices. That degree of access exceeds anything a single doctor has ever held.

The central question now is who owns the version of us that AI creates from our data. The companies that answer it convincingly will earn something more valuable than engagement: permission to enter the most private relationship in technology – the one between a person, their body and the future they are trying to protect.