Large language models (LLMs) and related generative artificial intelligence (AI) systems are rapidly entering clinical workflows, including in gastroenterology and hepatology, where text-heavy documentation, guideline-driven care, and high-volume patient messaging create a strong demand for decision support and automation. Deployment raises distinctive risks: hallucinations and unsafe recommendations, automation bias, privacy and confidentiality threats, inequitable care, intellectual property and licensing uncertainty, and unclear allocation of responsibility across clinicians, institutions, and vendors. Regulators are increasingly emphasizing a lifecycle approach to trustworthy AI, including validation, risk management, human oversight, cybersecurity, monitoring, and transparency. Frameworks for healthcare protection regulate how health information may be shared with external model providers and how outputs should be logged, audited, and retained. This review synthesizes practical guidance for responsible LLM use, organized around (1) tiered use cases and risk stratification; (2) ethical principles (beneficence, non-maleficence, autonomy, justice, and accountability); (3) core legal and regulatory considerations; and (4) operational governance, evaluation, and monitoring strategies. Actionable checklists to support institutional adoption, use, and transparency are provided. Responsible use requires aligning LLM capabilities with risk, preventing inappropriate data sharing, validating performance in representative populations, ensuring human oversight and clear accountability, mitigating harm, and maintaining an evidence-based, continuously monitored deployment lifecycle.
Sahil Khanna (Wed,) studied this question.