【Law vs. Medicine】Legal Lecture – Legal Liability and Risk Management of AI-Assisted Diagnosis in Clinical Decision-Making
Overview
On August 4, 2026, the Taipei Medical Association and the Chung-Hua Institution for Legal Research (the Foundation) held the second medical-law seminar in the “Law vs. Medicine” lunchtime series. The Foundation is sincerely grateful to the Chinese Society of Law for its support as co-organizer.
The topic was “Legal Liability and Risk Management of AI-Assisted Diagnosis in Clinical Decision-Making.” Professor Lee Chung-Hsi of Taipei Medical University was invited as the speaker. The Foundation was deeply honored that Dr. Chen Hsiu-Sheng, Deputy Superintendent of Taipei City Hospital, graciously agreed to serve as discussant. The moderator was Dr. Chou Hsien-chang of the Taipei Medical Association.
AI is increasingly used in clinical settings, from image interpretation and clinical decision-making to medical record writing. Several questions are of shared concern to the medical and legal communities:
- what happens when an AI suggestion is wrong;
- what happens when a physician does not adopt a correct AI suggestion;
- how hospitals should manage AI once it is introduced.
Opening Remarks
Dr. Hung De-jen, President of the Taipei Medical Association, thanked guests from the legal and medical communities for their long-standing support, which has enabled the two sides to keep exchanging new knowledge in healthcare and law. Dr. Chou Nai-Kuan, Chairperson of the Foundation, introduced the Festschrift in Memory of Justice Huang Mao-Zong, which President Wu Ming-Tzu had brought to the event. He also spoke about the links among law, taxation and AI as research frontiers. Wu Ming-Tzu, President of the Foundation, thanked the medical community for its support. She also expressed her admiration for the many medical professionals who have entered the field of law and bring clinical and legal perspectives together.
In his opening, moderator Dr. Chou Hsien-chang noted that the healthcare system faces challenges such as nursing shortages and heavy workloads for young physicians. Some young nurses, he said, even ask hospitals during job interviews whether they use generative AI to ease shift handovers. This shows that AI is already a real need on the clinical front line, and that the legal liability questions it raises need to be clarified.
Key Points of the Lecture
Starting from Article 82 of the Medical Care Act, Professor Lee compared U.S. law and then discussed Taiwan’s regulatory framework and hospital management strategies. He made four key points.
1. AI does not change the framework of Article 82 of the Medical Care Act
Under Article 82 of the Medical Care Act as amended in 2018, medical personnel are liable for damages only if they “breach the duty of care necessary in medical practice” and also “exceed the scope of reasonable clinical professional discretion.” In making this judgment, the medical practice, medical standards, medical facilities, working conditions and urgency of the situation at that time and place must be considered. Professor Lee noted that AI does not change this framework, but it does change the material and evidence used within it.
| Aspect | Requirement | Key Consideration |
|---|---|---|
| Objective conduct | Breach of the duty of care necessary in medical practice | Whether the conduct met the medical practice or standards of that time and place |
| Professional discretion | Exceeding the scope of reasonable clinical professional discretion | The medical facilities, working conditions and urgency at the time |
| Conclusion | Both requirements met | Only then is medical negligence established |
2. The scope of clinical discretion may gradually narrow
At this stage, AI remains an assistive tool. A physician who declines an AI suggestion based on clinical judgment, and records the reasons in the medical record, still acts within reasonable discretion. Professor Lee cautioned, however, that if AI becomes standard medical practice in the future, the discretion to “not use AI” or to “reject AI without reason” may be narrowed. Whether or not physicians adopt AI suggestions, they should therefore record their clinical reasoning in the medical record.
Different types of medical AI differ in how they are regulated and in the risks they carry:
| Technology Type | Main Clinical Functions | TFDA Review | Main Risks | Human Oversight Requirement |
|---|---|---|---|---|
| Software as a Medical Device (SaMD) | Image interpretation, ECG interpretation, surgical navigation | Must be registered and obtain a medical device license | Automation bias, black-box effect, model drift | Physicians must personally double-check; AI may not replace personal diagnosis |
| Generative AI | Draft medical records, administrative workflows, patient education materials | Not a medical device; managed under Ministry of Health and Welfare administrative guidelines | Hallucination, fabricated information, data leakage | Medical records must be reviewed, revised and signed by a physician |
| Autonomous AI agents | Automatically dispatching orders or adjusting medication doses once authorized | Not yet covered by the Ministry’s guidelines; no clear regulation | Escaping human control, automatically executing wrong instructions | Full automation is not advisable at this stage; key interventions require human approval |
3. Lessons from U.S. law
Professor Lee introduced the analysis of eight scenarios of physician liability by U.S. scholars Price, Gerke and Cohen. Under current standards of care, a physician is generally liable only when deviating from standard care and harm results. He also noted that some U.S. states have passed laws requiring disclosure to patients when AI is used in medical care, and that the scope of the duty to inform may expand accordingly.
4. Taiwan’s regulatory framework and hospital management
The Artificial Intelligence Basic Act passed its third reading in December 2025. It sets out basic principles including human autonomy, privacy protection, cybersecurity and safety, transparency and explainability, and accountability. The Ministry of Digital Affairs has also issued an “AI Risk Classification Framework.” In May 2026, the Ministry of Health and Welfare issued administrative guidelines on the use of generative AI by medical institutions, which emphasize that humans make the final decision.
Professor Lee advised hospitals introducing AI to build systems covering:
- governance;
- pre-adoption assessment;
- informed consent and disclosure;
- data governance;
- continuous monitoring;
- education and training.
He also advised them to observe four red lines:
- Do not enter identifiable sensitive data into public AI tools.
- Do not let AI replace a physician’s final decision.
- Do not use AI beyond the purposes approved by the Taiwan Food and Drug Administration.
- Do not fail to record the reasons for adopting or rejecting AI suggestions.
In clinical practice, the following table may serve as a reference:
| Clinical Situation | Recommended Practice | Practice to Avoid |
|---|---|---|
| Diagnosis and interpretation | Treat AI as an assistive reminder; have physicians personally double-check all images and data | Signing off AI reports marked “no abnormality” directly out of fatigue or heavy workload |
| Medical records and documents | Check AI-generated records word by word before signing; record reasons and keep records when departing from AI suggestions | Adopting AI-generated records without review, or letting AI fill in missing medical history |
| Doctor–patient communication | Explain the use of AI as appropriate; obtain patient consent before audio or video recording | Concealing the use of AI, or installing recording equipment in hidden places |
| Vendor contracts | Use official contract templates and require cybersecurity, warranty and monitoring clauses | Accepting vendors’ one-sided disclaimers without legal and cybersecurity review |
Discussant’s Remarks
From a hospital management perspective, Dr. Chen Hsiu-Sheng explained how liability is allocated when AI makes an error:
- physicians bear liability for medical negligence;
- AI developers may bear product liability;
- medical institutions may bear joint liability for management failures.
Using clinical scenarios, he also highlighted the risks AI may bring:
- “automation bias,” in which physicians rely too heavily on AI and give up independent judgment;
- the “black-box effect,” in which the AI’s reasoning cannot be explained;
- “model drift,” in which accuracy declines as conditions change.
In his view, the core of legal disputes often lies not in how intelligent AI is, but in how physicians and medical institutions use it. A physician’s independent judgment and a hospital’s continuous quality monitoring of AI are fundamental to avoiding legal risk.
Conclusion
AI is an assistant, not a decision-maker. Professor Lee reminded physicians to record their clinical reasoning whether they adopt or reject AI suggestions. He also urged them to watch how the duty to inform develops, and to be aware of the risks that “not using AI” may pose in the future. Understanding what the law requires allows physicians to protect patient safety, and themselves, as they adopt new technology.
For more on medical records and AI in medicine, please see these related seminars in the “Law vs. Medicine” series:
- Strengthening Medical Record Management and Legal Awareness (May 13, 2025; Chinese only)
- 【20261109 Law vs. Medicine: Legal Risks Every Primary Care Physician Should Know】
This English version was prepared with the assistance of AI. Please refer to the Chinese text for accuracy.
Zusammenfassung
【20260804 Justiz vs. Medizin: Haftung und Risikomanagement bei KI-gestützter Diagnose】
Am 4. August 2026 veranstalteten die Taipei Medical Association und die Chung-Hua Institution for Legal Research (die Stiftung) das zweite medizinrechtliche Mittagsseminar der Reihe „Justiz vs. Medizin“. Die Stiftung dankt der Chinese Society of Law herzlich für ihre Unterstützung als Mitveranstalterin. Es referierte Prof. Lee Chung-Hsi von der Taipei Medical University; als Diskutant wirkte freundlicherweise Dr. Chen Hsiu-Sheng, stellvertretender Superintendent des Taipei City Hospital, mit.
Prof. Lee erläuterte, dass KI den Prüfungsmaßstab für ärztliche Fahrlässigkeit nach Art. 82 des Medical Care Act nicht verändere, wohl aber die Beweisführung. Sollte KI künftig zum medizinischen Standard werden, könne sich der ärztliche Ermessensspielraum verengen. Ärztinnen und Ärzte sollten daher ihre klinischen Überlegungen dokumentieren, gleich ob sie KI-Empfehlungen folgen oder nicht. Dr. Chen erläuterte die Haftungsverteilung zwischen Ärzteschaft, KI-Entwicklern und medizinischen Einrichtungen. Er betonte, dass das unabhängige ärztliche Urteil und eine fortlaufende Qualitätskontrolle der KI durch die Krankenhäuser entscheidend seien.
Diese Zusammenfassung wurde mit Unterstützung von KI erstellt. Maßgeblich ist der chinesische Text.
要約
【20260804 司法vs.医療:AI支援診断の法的責任とリスク管理】
2026年8月4日、台北市医師公会と財団法人中華法学研究院(以下「本法人」)は、「司法vs.医療ランチタイム講座」医事法シリーズ第2回を開催しました。協力団体としてご支援いただいた中華法学会に、心より感謝申し上げます。台北医学大学の李崇僖教授を講師にお招きし、台北市立聯合病院の陳修聖副総院長にコメンテーターをお引き受けいただきました。
李教授は、AIは医療法第82条に基づく医療過失の判断枠組みを変えるものではないが、その判断に用いる材料や立証のあり方を変えると説明しました。将来AIが医療慣行として定着すれば、臨床上の裁量の幅が狭まる可能性もあるとしました。そのうえで、AIの提案を採用するか否かにかかわらず、臨床上の判断理由を診療録に記録するよう呼びかけました。陳副総院長は、医師、AI開発者、医療機関の間の責任分担について説明しました。また、医師の独立した判断と、医療機関によるAIの継続的な品質管理こそが、法的リスクを避ける要であると強調しました。
本要約は AI の支援を受けて作成したものです。正確な内容については中国語の本文をご参照ください。