Taiwanese hospitals are updating informed-consent packets and clinician workflows to align with Ministry of Health and Welfare guidelines on generative artificial intelligence, as outpatient clinics begin testing documentation assistants that summarize visits for physicians.
What the MOHW rules require
Guidelines issued in May 2026 tell medical institutions that deploy or plan generative AI to establish governance boards, document model purposes, assess patient-safety risks, protect personal data, disclose when AI contributes to patient communication, and retain audit logs. The rules do not ban clinical use; they demand traceability comparable to other high-risk hospital IT systems.
Legal analysts note the standards now serve as benchmarks under Taiwan’s Artificial Intelligence Basic Act, giving regulators a template while sector-specific penalties are still being drafted ahead of a January 2028 statutory deadline.
Why consent forms are changing
Consent is the patient-facing hinge. Hospitals must explain whether AI drafts chart notes, suggests billing codes or translates patient messages. Taipei medical centers told staff this month to add plain-language paragraphs describing human review requirements and opt-out paths where clinically feasible.
Privacy officers worry about models trained on cloud servers outside Taiwan. Contracts now require vendors to specify data residency, deletion schedules and whether prompts could be reused for model training.
Pilot projects on the ground
Several district hospitals are piloting ambient scribes that listen to encounters and produce draft summaries. Nurses review outputs before they enter electronic medical records; physicians must click to accept each paragraph, creating a log for malpractice insurers.
Radiology departments are farther along with computer-aided detection, but generative text tools introduce hallucination risks—fabricated symptoms or allergies—that clinicians must catch before orders go live.
Patient advocates weigh in
Consumer groups want hotlines for reporting AI errors and clear signage in waiting rooms. They argue transparency builds trust more than burying disclosures in multi-page forms.
For elderly patients who skip digital portals, hospitals are printing QR codes linking to audio explanations in Taiwanese and Hakka where budgets allow.
Connection to national cyber policy
The National Institute of Cybersecurity’s plan to use agentic AI in government drills underscores that health data remains a prime target. Hospitals aligning with MOHW guidance are simultaneously hardening APIs that connect scribe vendors to on-prem records.
Clinicians should expect more training hours this autumn as compliance officers roll out checklists. Patients should read new consent sections carefully: the technology can save documentation time, but responsibility for care decisions still rests with licensed professionals.