On 21 October 2026, at the Abu Dhabi Future Health Summit, the Department of Health – Abu Dhabi (DoH) published its AI Clinician Framework. WAM, the UAE state news agency, reports that DoH calls it a world-first regulatory framework. It sets out how AI clinician systems could eventually be authorised to perform clinical tasks now reserved for licensed clinicians. The framework governs research within Abu Dhabi's "living lab" model. It does not yet authorise anything. DoH says future policies and standards will define the permitted activities, the evidence required and the evaluation benchmarks.
That sequencing is the framework's main strength.
The strongest case for caution
The case for strict rules is serious. A clinical AI that errs does so at scale, and patients cannot easily judge a system's competence for themselves. Regulators in most jurisdictions have responded by treating clinical software as a medical device or by leaving the licensed human in the loop, so that no machine holds clinical authority of its own. A framework that lets a machine perform tasks reserved for licensed clinicians could look like a step too far.
That reading overstates what DoH has published. According to the announcement, competence is judged task by task, and broader scope requires stronger evidence. Physician oversight, accountability and referral to a physician apply at every level of scope. The framework does not grant a general licence. It describes a ladder, and a system has to earn each rung with evidence.
Why task-level competence is the right unit
Most regulation of clinical technology judges a product as a whole. Clinical risk, though, depends on the task. Triaging a symptom checker's output, drafting a discharge summary, interpreting a retinal scan and prescribing a drug carry very different risks. A regime that attaches evidence to the task lets low-risk uses move first and keeps high-risk ones waiting for proof. That is the proportionate, evidence-based approach this publication favours. It avoids both a blanket ban and a blanket permission.
The design also keeps accountability with a human. Because physician oversight and referral apply at every level, the framework avoids the hardest unresolved question in AI liability, namely who answers when an autonomous system harms a patient. For now the answer stays a named physician.
What is still missing
The framework is a skeleton. The benchmarks, evidence thresholds and permitted activities are all deferred to later DoH standards. Innovators cannot yet plan around them. If the standards arrive slowly or set evidence bars that only large vendors can meet, the "living lab" could become a showcase rather than a pathway. Three things will decide that:
- Published benchmarks: DoH should release evaluation criteria in draft and take comment, so developers and clinicians can challenge them before they harden.
- Transparent results: Evidence from the living lab should be public and comparable. Otherwise other regulators cannot learn from it, and the "world-first" claim is just a slogan.
- A defined step from research to authorisation: Developers need to know what evidence moves a system out of research into permitted use, and on what timeline.
Fit with the existing UAE stack
The framework lands in a layered system. Per the International Bar Association's overview, Abu Dhabi issued the region's first emirate-level healthcare AI policy on 30 April 2018. It then added a Responsible AI Standard in 2025, and federal rules already cover health data and medical software. Federal Law No. 2 of 2019 on ICT in health fields applies across the UAE, free zones included. MoHAP's page on the law sets out its scope. The IBA overview adds that the law requires health data to be stored and processed locally, subject to exceptions. Federal Decree-Law No. 38 of 2024 is the main law for software as a medical device. The federal UAE government portal also lists the legislation and policies for ICT in the health sector.
That layering is a mixed picture for innovators. A research-stage framework at emirate level is easy to enter. Commercial deployment would still have to satisfy federal medical-product registration and data-localisation rules. DoH, MoHAP and the free-zone regulators will need to line up their standards, or developers will meet the real friction at the point of authorisation rather than in research.
Assessment
DoH has made a defensible choice. It publishes the principles first and the technical standards later, and it keeps a physician accountable at every rung. A light-touch, research-first approach fits a technology whose real-world performance is still being measured. The "world-first" label is DoH's own, and the framework text was not available for this analysis. The details here come from the announcement and press coverage. The test is whether the promised standards are specific, public and timely. If they are, Abu Dhabi will have given the field a usable template for graduated clinical authority. If they are not, the framework will be an announcement without a pathway.