Who Bears Responsibility When AI Prescribes Psychotropic Medications?

As medical artificial intelligence becomes deeply involved in diagnosis and prescription, a gap in legal liability is being revealed. Particularly when dealing with high-risk medications like psychotropic drugs, this ambiguity raises fundamental questions about regulation and patient safety. Between an AI's recommendation and a human doctor's final judgment, where should the line of responsibility be drawn?

Current Status: Present Regulations and International Differences

Currently, neither the U.S. FDA nor the Korean Ministry of Food and Drug Safety (MFDS) has established independent, specific regulations solely for 'AI-based psychotropic medication prescribing.' These systems are evaluated under general guidelines for AI-based medical devices and clinical decision support. However, AI related to mental disorders is mostly classified as high-risk and approved as 'prescription-only devices,' requiring that actual prescriptions must undergo the final judgment of medical professionals according to each country's narcotics control laws and regulations. There are no confirmed cases of explicit approval for a process where AI automatically prescribes psychotropic drugs independently without physician intervention.

Regulatory approaches to psychotropic medications vary distinctly by country. According to academic research, a study comparing the prevalence of psychotropic medication use among adolescents in the United States, the Netherlands, and Germany highlights differences in regulatory restrictions and reimbursement policies. A study internationally comparing the approved information for pediatric depression medications also points out discrepancies in national approvals and prescribing guidelines. A policy report from the Korean Ministry of Food and Drug Safety provides a detailed analysis of these differences in national prescription monitoring systems and approval standards.

Analysis: The Liability Gap and the Role of Societal Consensus

Legal frameworks are still designed around the traditional doctor-patient relationship. When AI is deeply involved in the treatment process, it becomes unclear who bears ultimate responsibility—the manufacturer, software developer, prescribing physician, or institution—if harm occurs due to algorithmic errors, data bias, or unexplainable decisions. This creates significant uncertainty from the perspectives of patient protection and litigation risk management.

Interestingly, these regulatory differences are difficult to explain purely based on medical evidence. Comparative studies analyze that social sentiment, cultural acceptance, and the structure of the healthcare system profoundly influence the strictness of regulation and prescribing patterns. In other words, whether to allow high-risk drug prescribing by AI is an issue that goes beyond simple technical safety assessment; it requires prior ethical and political consensus on how society defines and distributes risk.

Practical Application: Questions to Consider When Implementing Systems

When healthcare institutions introduce AI for psychotropic medication prescribing support or when policymakers prepare related guidelines, it is necessary to ask several clear questions. First, whether the system's output is a 'recommendation' or an 'instruction' must be clearly defined at the protocol level. Second, it should be verified through legal consultation how the country's narcotics control laws and AI guidelines are linked, particularly identifying the mandatory points of physician intervention. Finally, whether a physician can understand and explain the rationale behind the system's decision (explainability) should be a core condition for technology adoption.

FAQ

Q: If side effects occur from psychotropic medications prescribed by AI, who can the patient sue? A: Under the current regulatory system, it is difficult to consider the AI itself as a legal entity. Therefore, patients can typically seek liability from the physician who ultimately approved and signed the prescription, or from the manufacturer if there was a defect in the software. However, specifying negligence within the AI's complex decision-making process is very difficult, potentially leading to legal gaps and disputes.

Q: Is a physician's final signature mandatory for AI psychotropic medication prescribing support systems in all countries? A: According to the current guidelines of the U.S. FDA and the Korean MFDS, AI related to psychotropic medications is classified as high-risk and approved as prescription-only devices. Therefore, officially, final review and approval by a physician are absolutely necessary. However, each country's narcotics control laws must be reviewed in detail, and how this principle will be applied as technology advances remains to be seen.

Q: If a patient's condition worsens because a physician did not follow the AI's prescribing recommendation, who is responsible? A: The physician is the ultimate decision-maker responsible for integrating all information, including the AI's recommendation, to make their own professional judgment. Therefore, if the physician refused the AI's recommendation based on reasonable grounds and that decision aligned with the standard of care at the time, the physician may be free from liability. Conversely, if ignoring a clear and appropriate warning from the AI is judged as negligence, liability may arise.

Conclusion

The issue of liability in AI medical prescribing, especially in the realm of psychotropic medications, is a typical case where regulation and jurisprudence cannot keep pace with the speed of technological advancement. The solution lies not simply in putting 'brakes' on the technology, but in designing a new model for distributing responsibility among physicians, algorithms, and institutions through societal consensus. The immediate course of action is clear: every institution implementing AI systems must first establish its own legal and ethical boundaries, prioritizing them over technical utility.

참고 자료

• 🛡️ Clinical Decision Support Software | FDA
• 🛡️ [Generative AI-Enabled Digital Mental Health Medical Devices](https://www.fda.gov/advisory-committees/digital-health- advisory-committee/november-6-2024-digital-health-advisory-committee-meeting-announcement)
• 🛡️ 식품의약품안전처 가이드라인 자료실
• 🛡️ A three-country comparison of psychotropic medication prevalence in youth
• 🛡️ 식품의약품안전처 연구관리시스템 - 연구보고서
• 🏛️ An International Comparison of the Information in the Regulatory-Approved Drug Labeling and Prescribing Guidelines for Pediatric Depression