Kenya’s medical regulator has issued a sweeping new directive on technology-based healthcare, warning practitioners that artificial intelligence is a support tool only and that they bear full professional responsibility for every treatment decision, including those assisted by AI.
The Kenya Medical Practitioners and Dentists Council (KMPDC) published the public notice outlining professional obligations for telemedicine, e-health, m-health, robotic medicine, wearable devices, and artificial intelligence, issued pursuant to the Code of Professional Conduct and Ethics, 7th Edition (2026).
The guidelines arrive at a critical moment. Kenya has rapidly become a testing ground for AI-assisted clinical care in primary settings, even as the legal and ethical framework around digital health remains contested.
The Council’s notice explicitly requires practitioners and health institutions to comply with several laws, including the Health Act, the Digital Health Act (No. 15 of 2023), the Medical Practitioners and Dentists Act, the Data Privacy Act, and the Civil Aviation Regulations governing unmanned aircraft systems.
“Recognise that artificial intelligence is a support tool only and does not replace human decision-making or clinical oversight,” the Council stated in the notice, adding that practitioners must “accept full professional responsibility for all treatment decisions and recommendations, including those assisted by artificial intelligence systems.”
The regulator also moved to close a loophole that has troubled patients and regulators alike. Practitioners are now explicitly barred from signing or permitting the use of their name, signature, registration, or licence number to indicate they provided a service or issued a prescription they did not actually provide.
The notice further prohibits prescribing for any person whose condition has not been properly ascertained, a measure targeting casual remote prescribing in some digital health services.
The Council’s intervention follows a landmark real-world evaluation of an AI clinical decision support system deployed across 16 primary care clinics in Kenya, with findings that attracted international attention in the global health AI community earlier this year.
The study, published in Nature Health and based on encounters between July and September 2024, examined an off-the-shelf GPT-4o tool integrated into electronic medical records at Penda Health clinics in Nairobi and Kiambu counties.
Clinicians, largely non-physician providers licensed to diagnose and prescribe independently, could click an “AI Consult” button during patient encounters to receive diagnostic and management suggestions.
The findings were mixed. The AI’s management guidance aligned with local Kenyan clinical guidelines in 99% of reviewed encounters, while reviewers found that the tool provided new diagnostic insights beyond what clinicians had documented in 94% of cases. Hallucinations were uncommon, at 3.4%.
However, a concerning pattern emerged when clinicians acted on the AI’s advice. Harmful AI outputs were adopted fully or partially 58% of the time, while beneficial AI outputs were adopted in only 22% of cases.
The AI produced actively harmful recommendations in 7.8% of encounters, with 67 of those recommendations making it into final patient records.
The researchers attributed the problem to confirmation bias rather than blind trust. Clinicians were more likely to accept AI recommendations that supported what they had already documented, including errors, while dismissing suggestions that challenged their initial decisions, even when the AI recommendations were correct.
“AI will not replace clinicians. But poorly designed AI will reshape how clinicians make mistakes,” noted one of the study’s discussants, Dr. Ethan Goh, in commentary that circulated widely among health informatics specialists.
The KMPDC notice also comes amid legal uncertainty surrounding Kenya’s digital health framework. The Digital Health Act was declared unconstitutional by the High Court in July 2024 in a judgment that disrupted the regulatory framework and left agencies seeking to establish their jurisdiction over virtual healthcare.
A subsequent petition filed by the Kenya Association of Radiologists accused the Council and other regulators of “regulatory inertia,” alleging that unlicensed AI-assisted teleradiology services were exporting patient images outside Kenya for reporting by unidentified foreign individuals without consent.
Against this backdrop, the KMPDC’s requirements that practitioners use only tools meeting high standards of data security and breach prevention, maintain accurate and complete records, and facilitate patients’ access to their medical records in portable formats appear aimed at reinforcing minimum safeguards.
The notice also requires practitioners to provide services only within their defined scope of practice, a provision that could affect informal remote consulting networks that have emerged across sub-Saharan Africa, where health workers frequently use personal phones to consult patients at their own expense.
For patients, the guidelines offer greater clarity on accountability in digital healthcare. Practitioners conducting telemedicine consultations must accurately identify themselves, properly assess patients’ conditions before prescribing and take responsibility for the medical advice they provide.
The requirement that patients be allowed to access their medical records in accessible and portable formats also strengthens their ability to obtain and retain personal health information.
For practitioners, however, the message is that using artificial intelligence does not shield them from professional liability. The Council has made clear that the AI consultation feature cannot replace clinical judgment or transfer responsibility to the technology.
“The question for LLM-based clinical decision support is not ‘whether the model is good enough,’” the study discussants concluded. “The real challenge is designing a robust human-AI interaction layer – explainability, trust – so that clinicians act on good advice and ignore bad advice”.
