Safe Deployment of a Generative AI Assistant for Traditional-Medicine Education: Defense-in-Depth Architecture, Domain-Calibrated Safety Evaluation, and a Proposed Minimum Safety Stack
Abstract
Abstract Deploying generative AI safely in traditional-medicine education requires more than model alignment: it demands layered architecture and a definition of ‘unsafe’ that reflects clinical and pedagogical standards, not only information-security criteria. We describe TLAS-YHCT, a Socratic chatbot for traditional-medicine students built around non-overridable guardrails, retrieval-augmented generation over a vetted clinical corpus, a standardised system prompt, and an independent LLM-as-judge verifier. A structured red-team of 206 attacks—constructed by domain-expert clinician-educators independent of system developers, scored blind by the same experts under a dual IT-security and clinical-education safety criterion—yielded a 100% safe-response rate versus 88.9% and 78.7% for two commercial baselines. Critically, several attack scenarios were judged safe by IT-security standards but clinically unsafe by clinical-education standards, and vice versa, demonstrating that safety definitions must be domain-calibrated. We propose a minimum safety stack—combining RAG, a standardised system prompt, and LLM-as-judge verification—as a transferable framework for AI safety in traditional-medicine and allied clinical-education settings.
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.