# How to Test UAE Healthcare AI Claims

The UAE has no shortage of healthcare AI announcements. The useful question for a buyer is narrower: what evidence shows that a system is working in a real service, and what remains a claim by the organisation deploying it?

A new [Emirates News Agency report](https://www.wam.ae/en/article/c1oidjs-algorithms-smart-technology-shape-uae-healthcares) published on 11 August describes AI spreading across diagnosis, hospital operations and patient services. The underlying programmes show genuine activity, but they sit at different levels of maturity. A contact-centre classifier, a clinical imaging model and a regulator's analytics platform should not be assessed with the same evidence.

This distinction matters for health CIOs, product teams and investors. An impressive demonstration can be useful. It is not a clinical outcome.

## What is actually in use

Emirates Health Services (EHS), the federal health provider, lists AI across clinical and administrative work. Its programme page says mammography algorithms supported breast-cancer diagnosis for **532 patients at four hospitals between 2019 and 2022**. EHS also reports that voice recognition reached **82 targeted hospitals and medical centres**, with **1,800 physicians trained** and **1,200 active doctors**.

Those figures establish scale better than a product launch does. EHS also reports that voice recognition increased documentation input by 83%, reached 97.6% clinical-documentation accuracy at Abdullah bin Umran Hospital and 90% at Dibba Hospital, and cut consultation waiting time from 30 minutes to 15 minutes. These remain [EHS-reported results](https://www.ehs.gov.ae/en/about-us/emerging-technology/artificial-intelligence), not an independent evaluation, but they identify users, sites and operational measures.

The provider expanded the catalogue at World Health Expo Dubai in February 2026. Its [official release](https://www.ehs.gov.ae/en/media-center/news/emirates-health-services-unveils-advanced-ai-powered-system-driving-a-comprehensive-healthcare) describes Derma AI for skin imaging, AI-supported low-dose CT screening for lung cancer, MediVision for safety monitoring, a clinical information bot, a blood-donation app, a concierge and workforce tools.

The wording needs care. The release says these systems were showcased and describes intended benefits. It does not publish sensitivity, specificity, false-negative rates, patient outcomes or external validation for every clinical model. A buyer should record the products as named initiatives, then ask which are routinely deployed and which remain exhibition-stage projects.

## Service AI and clinical AI need different tests

Dubai Health Authority's Risk Radar is a useful contrast. The system analyses customer-service calls using sentiment, tone, contact frequency, behavioural patterns, surveys and CRM data. DHA says cases that need attention go to an operations team and supervisors can monitor trends through a live dashboard.

DHA reports [more than an 81% reduction in calls classified as negative over eight months](https://www.dha.gov.ae/en/media/news/993). That is a concrete reporting period and result, but the metric is still defined by the authority's own AI classification. It does not tell readers the starting number of negative calls, the model's error rate, whether call mix changed, or whether independent customer-satisfaction measures moved in the same direction.

This is not a reason to dismiss the system. It is a reason to ask for the denominator, the baseline and a human-reviewed sample. The operational risk is mostly customer service and escalation. A model that reads a lung CT or influences an urgent-care decision needs a much higher clinical bar: representative test data, subgroup performance, comparator results, workflow monitoring and a documented route for clinician override.

![Five-step evidence ladder for UAE healthcare AI, moving from announcement through operational deployment and measured results to clinical validation and governed scale, with UAE examples and buyer questions.](https://cdn.hashnode.com/uploads/covers/60ecf4a0fc37a15ec15655e8/5471fb6d-328a-45a8-8ca7-f0a3329b591f.png)

*Sources: [EHS AI programme](https://www.ehs.gov.ae/en/about-us/emerging-technology/artificial-intelligence) (updated 18 March 2025), [DHA Risk Radar](https://www.dha.gov.ae/en/media/news/993) (11 February 2026), [Abu Dhabi DoH AI policy](https://www.doh.gov.ae/en/news/the-department-of-health-becomes-region) (27 May 2018), and [DoH-Microsoft MoU](https://www.doh.gov.ae/en/news/DoH-to-Integrate-AI-powered-Technologies) (16 October 2024). SultanByte original infographic.*

## The UAE is not one health-AI market

Regulatory scope differs inside the country. Abu Dhabi's Department of Health published its healthcare AI policy in May 2018. The department says it applies to DoH-licensed providers, Abu Dhabi pharmaceutical manufacturers, insurers, licensed researchers involving human subjects, and organisations using Abu Dhabi population or patient data for AI. That is a broad scope, covering more than hospitals.

Abu Dhabi has also used partnerships to define future capability. An October 2024 [DoH-Microsoft memorandum](https://www.doh.gov.ae/en/news/DoH-to-Integrate-AI-powered-Technologies) names three areas: patient care, clinical trials and research, and regulator decision-making. A memorandum records intent and workstreams. Procurement teams should not cite it as proof that every promised model is deployed.

At federal level, the Ministry of Health and Prevention's [AI Office](https://mohap.gov.ae/en/w/artificial-intelligence-office) lists dashboards, predictive models, process automation and chatbots. Its projects include health-crisis management, organ donation, fraud detection for health certificates, engineering-plan review and forecasts for births, deaths and morbidity. These are mostly regulatory and administrative uses rather than bedside diagnosis.

Dubai's Risk Radar sits in customer operations, while EHS reports clinical, documentation and forecasting systems across its facilities. The country therefore has several buyers, regulators and data controllers. A product cleared for one workflow or emirate should not be assumed to have approval, data access or clinical acceptance elsewhere.

## A practical evidence request for buyers

Start with the exact decision the model influences. Ask whether it recommends, prioritises, drafts, flags or acts. Then map the failure: a missed service complaint is different from a missed malignancy.

For an operational product, request the deployment date, number of active sites, number of trained and active users, uptime, override rate and incident history. A cumulative training count can hide weak current adoption. A percentage improvement is hard to judge without its denominator and baseline period.

For clinical AI, add the intended patient population, data provenance, external validation, subgroup results, false-positive and false-negative rates, comparison with current practice, and post-deployment drift monitoring. Performance from one hospital may not transfer cleanly to another population, scanner or workflow.

Governance should name an accountable clinical owner and a technical owner. It should also cover access to patient data, audit logs, model changes, human override, incident reporting and retirement criteria. The [World Health Organization's guidance](https://www.who.int/publications/i/item/9789240029200) puts ethics, human rights and accountability around the full lifecycle, not just model development.

Investors need a similar discipline. Regulatory access, hospital integration and repeat use are stronger signals than the number of AI features in a presentation. Evidence of a paid deployment is useful; evidence of sustained use and measured benefit is better. Clinical validation remains a separate hurdle.

## From a catalogue to accountable services

The UAE's healthcare organisations have moved beyond a single AI pilot. EHS reports multi-site use, DHA has published an eight-month service metric, and Abu Dhabi has had a sector policy since 2018. The public evidence is still uneven. Some initiatives have named sites and measures; others are showcases, planned collaborations or institution-reported claims without independent validation.

That is the sensible reading of the current wave. Buyers should reward deployments that disclose scope, baselines, failure modes and monitoring. The next useful releases will publish denominators and independent clinical results, not longer lists of AI products.

*Cover: original SultanByte visual.*