Ministry of Health Approves Guidelines for Using AI Tools: Outputs Are Supportive, Not Definitive
- It must not be considered a definitive reference, an approved clinical recommendation, or a substitute for evaluation by a physician or healthcare specialist.
- It must not be used with actual patient data or confidential or operational data belonging to the Ministry unless officially approved.
The Undersecretary of the Ministry of Health, Sheikh Dr. Salman Al-Sabah, issued an administrative decision approving and publishing the Practical Guide for Using Linguistic Artificial Intelligence Tools and Large Language Models in Healthcare Practice. This guide serves as a regulatory framework for staff in healthcare facilities affiliated with the Ministry of Health, aiming to balance the benefits of these technologies with the protection of patient privacy, the safety of practice, and professional accountability.
The guide establishes a structured framework for using linguistic AI tools, including smart chatbots, summarization, translation, and drafting assistance tools such as ChatGPT, Claude, Gemini, OpenEvidence, and others, within the health, administrative, educational, and research contexts associated with the Ministry of Health, whether used inside or outside Ministry facilities.
The decision emphasized that the guide’s objective is not to prevent the use of AI tools, but to enable their safe and responsible use. It stressed that AI outputs remain supportive and non-definitive, do not constitute a substitute for professional judgment or medical responsibility, and must not be regarded as a definitive reference, an approved clinical recommendation, or a replacement for evaluation by a physician or healthcare specialist.
One of the guide’s key pillars emphasizes protecting patient data confidentiality and health information. It prohibits entering, uploading, or copying identifiable or confidential data—including patient names, civil ID numbers, medical file numbers, images, audio recordings, laboratory results, radiology reports, discharge summaries, and screenshots—into any linguistic tool, cloud service, or platform not officially approved by the relevant authorities within the Ministry.
The guide also stipulates that linguistic AI tools or associated cloud services must not be used with actual patient data or confidential or operational data belonging to the Ministry unless officially approved, in accordance with data protection, information security, cloud computing, patient confidentiality, and applicable technical and legal regulations in the State of Kuwait. It affirmed that a tool’s popularity or public availability does not constitute sufficient grounds for its use in patient-related work or Ministry data operations.
The guide sets clear boundaries for clinical use, prohibiting reliance on linguistic AI outputs as the sole basis for diagnosis, treatment, prescription, dosage calculation, test interpretation, admission/discharge decisions, referrals, triage, or predicting disease severity. Conversely, it permits their use as aids for learning, structuring thought, exploring possibilities, and summarizing information, provided that outputs and their sources are rigorously verified and reviewed by the responsible specialist before professional or clinical use.
The guide classifies uses according to risk level into three categories: low-risk uses include general drafting, translation, summarization, and preparation of non-confidential educational or administrative materials with human review; medium-risk uses include summarizing guidelines, organizing differential diagnoses, and reviewing general pharmaceutical information with source verification; and high-risk uses, where AI must not be relied upon for decision-making, include final diagnosis, treatment selection, prescription or dosage determination, clinical triage, and decisions concerning specific patients.
The resolution emphasized the principle of human professional responsibility, stipulating that physicians, healthcare specialists, or employees remain professionally accountable for any decision, recommendation, documentation, communication, or content they adopt within the scope of their work. The use of artificial intelligence (AI) tools does not exempt individuals from professional, legal, or ethical liability.
Regarding information verification, the guideline underscored the necessity of reviewing AI-generated outputs for their sources, logic, relevance to the local context, and compliance with the policies and directives of the Ministry of Health. It stressed that the quality of phrasing or the confidence of the style does not constitute evidence of factual accuracy, and that outputs must not be accepted merely because they are presented in a persuasive or well-organized manner.
The guideline paid special attention to Arabic content and translation, affirming that linguistic correctness does not necessarily imply the accuracy of scientific or clinical content. Any Arabic content intended for patients or the public that is generated, translated, or simplified using these tools must undergo review and approval by a responsible healthcare specialist. This ensures clinical accuracy, translation integrity, cultural appropriateness, and suitability for the health literacy level of the target audience.
Furthermore, the guideline prohibited health facilities, departments, and units from launching, using, or promoting language-based tools aimed at patients or the public to provide health advice, treatment information, or triage and clinical referral services under the name of the Ministry of Health or any of its facilities, unless a formal evaluation and approval mechanism is established. This mechanism must consider clinical, legal, and technical aspects, as well as information security, data privacy, patient safety, and performance monitoring of the tool.
The resolution authorized the Ministry, when necessary, to form a technical team or committee to evaluate AI tools and models proposed for use in official work or integration with the Ministry’s systems. The evaluation must take into account the level of risk, data types, privacy, cybersecurity, performance quality, verifiability of outputs, relevance to the local context, and mechanisms for monitoring and handling errors and incidents.
The guideline also included provisions for documentation and disclosure, affirming that no text generated or modified by AI tools may be copied into medical records or official reports without being read, reviewed, corrected, and approved by the responsible specialist.
The resolution further mandated the dissemination of awareness and training materials for staff regarding the safe and responsible use of these tools. It also called for the availability of public awareness materials explaining the limitations of AI in health-related topics, including its potential to generate inaccurate, incomplete, or incorrect information, and clarifying that AI does not replace consultation with a physician or healthcare specialist, nor does it substitute for consulting approved medical sources.
In the context of enhancing governance and follow-up, the guideline obliged health facilities to report any incident, error, or suspected breach of patient privacy or unsafe use of language-based AI tools. This includes the unauthorized entry of patient or confidential data into unapproved tools, or the use of unverified outputs that may compromise patient safety or the quality of care.
Thus, the resolution establishes a comprehensive regulatory framework for the responsible use of language-based AI in healthcare practice, balancing the benefits of modern technologies with the reinforcement of human accountability, data and privacy protection, patient safety, and information verification. The guideline is subject to periodic review and updates, or as needed, to keep pace with the rapid scientific, regulatory, and technological developments in this field.
