Taiwan MOHW Targets AI Agent Governance in Upcoming Medical Guidelines

Taiwan's Ministry of Health and Welfare (MOHW) issued generative AI application guidelines for healthcare institutions in May, but autonomous AI Agents capable of independent decision-making and action were excluded. Department of Medical Affairs Director-General Liu Yueh-ping stated at a hospital CIO conference that future guidelines will address AI Agents, centering on authorization boundaries, human intervention, execution logs, emergency halts, and accountability, though release timelines remain unannounced. Liu emphasized that healthcare AI governance begins with comprehensive inventory tracking across commercial tools, internal models, RAG setups, departmental shadow AI, and system updates. Institutions must identify data flows, vendors, and responsible personnel. For high-impact tools, hospitals are advised to perform internal validations such as silent runs or limited pilots, followed by continuous operational monitoring, human oversight, and defined decommissioning procedures.

Key takeaways

• MOHW's upcoming medical AI guidance will cover autonomous AI Agents, focusing on authorization scopes, human takeover, action logs, emergency stops, and liability.
• The May generative AI guidance provides administrative recommendations structured around five principles, including risk ownership, data protection, and manual verification.
• Healthcare AI governance starts with inventorying commercial tools, internal models, RAG, shadow AI, and system updates to track data flows and accountability.
• Model deployment requires clinical validation through silent trials or limited pilots, paired with ongoing monitoring and exit mechanisms post-deployment.