§ Medical & healthcare · AI Basic Act
Medical: AI Basic Act compliance & Korean LLM choice
Using an LLM to summarize records or answer patient inquiries is a different regulatory story from letting it inform clinical judgment. Decide which one you are doing, and record why.
Why this sector is high-impact
Medical is named among high-impact fields too. But there is a second layer here, separate from the AI Basic Act: software that supports diagnosis or clinical judgment may fall under medical-device regulation, which carries a far heavier process. That is a question to settle before choosing a model, not after.
Key duties
- ✓First determine medical-device applicability (if it supports diagnosis/judgment)
- ✓Impact assessment for high-impact AI
- ✓Fix where patient data is processed, cross-border scope, retention period
- ✓Retain safety and decision records
✦ Model-choice angle
Cross-border review bites first with patient data, which makes self-hostability effectively the top constraint. Commercial open weights (A.X 4.0) as the default candidate, with EXAONE considered after a license check, is the easier order to defend. Korean clinical notes are dense with abbreviations and non-standard forms, so general benchmark scores transfer poorly — measuring hallucination and omission on your own record samples is the only citable evidence.
Related
References
General information, not legal advice. Verify applicability and duties with official MSIT guidance and legal/compliance counsel.