§ AI Basic Act
Korea's AI Basic Act & Korean LLM adoption — model choice is compliance
Korea's AI Basic Act (effective 2026-01-22) places duties on companies that adopt third-party LLMs, not just model builders. The documented rationale for which model you chose becomes a core artifact for impact assessment and recordkeeping.
›Who it applies to
Not only model developers but companies that adopt/operate third-party LLMs are in scope under discussion, with heavier treatment for high-impact domains (lending, hiring, healthcare). Scope varies by sector/use — verify with official guidance and legal counsel.
›Core duties (gist)
Effort to conduct impact assessments for high-impact AI, transparency/labeling duties, and safety-related record retention are the main axes. Public bodies are directed to prefer products that have undergone impact assessment.
›Why model rationale is the artifact
Documenting 'why this model' on cost, performance, Korean quality, and failure patterns yields evidence citable in assessments, procurement, and audits. Vendor-published numbers alone are weak on independence/reproducibility.
›Penalties / preference
Penalties for violations are defined (commentary cites up to ~₩30M); the grace period is described as preparation deferral, not duty deferral. Public procurement prefers assessed products.
✦ How KoreanLLM helps
- ✓Independent, vendor-funding-free evaluation of Korean LLM candidates → a model-selection evidence table
- ✓Transparent labeling of which scores are self-reproduced vs vendor-cited (audit-ready)
- ✓A one-page conclusion you can attach to assessments and procurement
Related
References
This page is general information, not legal advice. Verify applicability, duties, and penalties with official MSIT guidance and legal counsel.