Can a clinical AI system make autonomous final medical decisions under the Draft AI Ethics Guidelines?

No. Table 1 explicitly requires meaningful human control and prohibits fully automated final decisions on matters significantly impacting human life or health.

What is the definition of an AI system provider under the August 2025 draft guidelines?

Definition point 7 defines an AI system provider as an entity that provides access to AI models and integrates them into specific products, services, or platforms.

What emergency mechanisms must an AI system provider maintain?

Table 1 mandates clear emergency procedures and technical controls to shut down an AI system quickly and safely in the event of an operational anomaly.

How does the Draft AI Ethics Guideline govern AI system providers?

The August 2025 Draft AI Ethics Guidelines require AI system providers to implement risk management, maintain transparency, ensure human control over high-impact decisions, and provide incident reporting channels.

The Draft Concept of Artificial Intelligence (AI) Ethics Guidelines released for public consultation by the Ministry of Communications. Furthermore, Digital in August 2025 defines AI system providers as a distinct subcategory of Sector Actors. Rather than establishing isolated statutory sections, the draft assigns four core responsibilities collectively to Sector Actors. These cover lifecycle risk management, system transparency, human-in-the-loop oversight, and incident reporting mechanisms. The document remains a public consultation draft and is not yet enacted law.

Who qualifies as an AI system provider

The draft identifies four Sector Actor categories: data providers, AI developers, AI system providers, and AI system operators (Definition point 3). Under Definition point 7, an AI system provider is any entity that provides access to AI models. Furthermore, Underlying infrastructure, integrating artificial intelligence into specific products, services, or platforms.

Healthcare software vendors that embed machine learning models into electronic medical record applications or clinical documentation workflows represent AI system providers. At the same time, The hospital or clinic operating the software acts as the system operator.

Four core responsibilities of Sector Actors (Chapter VIII)

Developing and operating AI systems in alignment with ethical values and principles;
Executing risk management across every stage of the AI lifecycle;
Mitigating risks through prevention, early identification, remediation, periodic audits, continuous oversight, and system refinement. and
Guaranteeing transparency and accountability for deployed systems.

Safeguards and risk categorization (Table 1 and Chapter VII)

Table 1 details specific technical safeguards applicable to AI system providers:

Technical documentation: Providing clear explanations of model architecture, training data characteristics, and operational boundaries.
Incident reporting and emergency stops: Establishing complaint and incident channels alongside emergency procedures to shut down system operations quickly and safely.
Human oversight (Human-in-the-loop): Ensuring AI systems remain subject to meaningful human control. Prohibiting automated final decisions in matters significantly affecting human life, health, or fundamental rights.

The draft stratifies AI into three risk tiers: unacceptable risk (prohibited), high risk (subject to strict governance requirements). In addition, Minimal or low risk (requiring transparency compliance).

Self-assessment and incident remediation (Chapter IX)

Sector authorities establish self-assessment frameworks evaluating bias detection, accountability, explainability, privacy protection, and societal impact. User grievances submitted to authorities are referred to the reported Sector Actor to trace root causes, apply system fixes, and submit formal remediation reports.

The draft proposes a two-year preparation window following formal promulgation (Chapter X). As a consultation text, it does not currently impose binding legal duties.

Legal basis

Draft Concept of Artificial Intelligence Ethics Guidelines (Ministry of Communications. Furthermore, Digital, public consultation draft August 2025, not yet in force): Definition points 3, 7, and 8. Chapters IV, V, VII.2 Table 1, VIII, IX, and X.