# Korea AI Basic Act Compliance Assessment: [Organisation or product name]

> Compliance assessment for the duties on generative AI, high-impact AI
> and high-performance AI under the Republic of Korea's Framework Act on
> the Development of Artificial Intelligence and the Establishment of a
> Foundation for Trust, often called the AI Basic Act (Act No. 20676 of
> 21 January 2025, in force since 22 January 2026, amended by Act No.
> 21311 of 20 January 2026), its Enforcement Decree and the guidelines
> of the Ministry of Science and ICT (MSIT). Article numbers follow the
> English translation published by the Korea Legislation Research
> Institute. Replace every `[bracketed]` value and every row marked
> *Example*. This is a documentation format, not legal advice; the rules
> are new and change often, so check the current official text and a
> qualified Korean counsel before relying on it.

<!--
Guidance: How to use this template
- Use one assessment per product or service that reaches users or the
  market in Korea, including services run from abroad: the Act reaches
  AI activities outside Korea that affect the Korean market or users.
- Fill the dated fact table in section 2 first, from the official
  sources (law.go.kr, the KLRI English service and MSIT notices). Every
  threshold in this template comes from that table, so a change there
  flows through.
- The Act uses its own roles: AI development business operators and AI
  utilisation business operators. They do not map one to one onto the EU
  AI Act's provider and deployer; record your role per system.
- Several duties are worded as "efforts" rather than strict obligations.
  Record both, and record which carry an administrative fine.
-->

## Document control

<!-- Guidance: Identify the product, the Korean counsel who reviewed it and the date the official sources were checked. -->

| Field | Value |
|---|---|
| Document ID | [KR-AIBA-001] |
| Product or service | [Name, version] |
| Version | [0.1] |
| Status | [Draft / In review / Approved / Superseded] |
| Owner | [@ai-compliance-lead] |
| Reviewers | [@kr-counsel], [@product-lead], [@privacy] |
| Official sources checked on | YYYY-MM-DD |
| Domestic representative | [Name, address in Korea / Not required] |
| Last updated | YYYY-MM-DD |
| Classification | [Internal / Confidential] |

### Approvals

| Role | Name | Decision | Date |
|---|---|---|---|
| Compliance lead | [@name] | [Approve / Reject] | YYYY-MM-DD |
| Korean counsel | [@name] | [Reviewed] | YYYY-MM-DD |
| Product owner | [@name] | [Accept actions] | YYYY-MM-DD |

## Revision history

<!-- Guidance: Re-check the fact table at least quarterly while the Enforcement Decree and guidelines settle. -->

| Version | Date | Author | Change summary |
|---|---|---|---|
| 0.1 | YYYY-MM-DD | [@name] | First draft |
| 1.0 | YYYY-MM-DD | [@name] | *Example:* Approved after counsel review |
| 1.1 | YYYY-MM-DD | [@name] | *Example:* Updated after MSIT guideline revision |

## 1. Scope and roles

<!-- Guidance: List each AI system that reaches Korean users and the role you play for it. Pitfall: forgetting features built on a third-party model; you can still be a utilisation business operator. -->

| ID | System | Korean users or market | Role under the Act | Generative AI? | Possible high-impact area |
|---|---|---|---|---|---|
| SYS-001 *(Example)* | Tutoring chatbot | Learners in Korea through an app | Utilisation business operator (built on a third-party model) | Yes | No |
| SYS-002 *(Example)* | Essay scoring for schools | Contracted by Korean schools | Development and utilisation business operator | No | Possibly: student assessment |
| SYS-003 *(Example)* | Avatar video generator | Teachers create lesson videos | Development business operator | Yes | No |

*Example organisation:* Pelmora Learning, a fictional education
technology company established outside Korea.

## 2. Dated fact table

<!-- Guidance: Record each fact with the date checked and the source. If a fact comes from a draft or a secondary source, say so. -->

| Fact | Value as checked | Source | Checked on |
|---|---|---|---|
| Entry into force | 22 January 2026 | Act addenda; KLRI | YYYY-MM-DD |
| Latest amendment | Act No. 21311, 20 January 2026 | KLRI English service | YYYY-MM-DD |
| Enforcement Decree | Reported finalised with five MSIT guidelines at entry into force | MSIT, law.go.kr | YYYY-MM-DD |
| High-performance AI compute level | Reported as 10^26 FLOP cumulative training compute | Enforcement Decree | YYYY-MM-DD |
| Domestic representative thresholds | Reported as total revenue above 1 trillion KRW, AI service revenue above 10 billion KRW, or average daily Korean users above 1 million, in the prior year | Enforcement Decree | YYYY-MM-DD |
| Maximum administrative fine | 30 million KRW (Article 43) | Act | YYYY-MM-DD |
| Fine grace period | MSIT stated a period of at least one year before fines | MSIT announcement | YYYY-MM-DD |

Thresholds marked "reported" were taken from law firm summaries of the
decree when this template was written. Confirm them against the decree
on law.go.kr.

## 3. Transparency duties (Article 31)

<!-- Guidance: Applies to generative AI and high-impact AI. Check the advance notice, the output label and the clear notice for realistic synthetic media separately. The guidelines allow lighter treatment for artistic and creative works; record when you rely on it. -->

| ID | Duty (summary) | Systems | How met | Status | Evidence |
|---|---|---|---|---|---|
| KR-T-001 *(Example)* | Tell users in advance that the product or service runs on generative or high-impact AI | SYS-001, SYS-002, SYS-003 | Onboarding notice and terms in Korean | Done | Screenshots |
| KR-T-002 *(Example)* | Indicate that outputs are produced by generative AI | SYS-001, SYS-003 | Visible label in chat; machine-readable metadata in files | Done | Label specification |
| KR-T-003 *(Example)* | Give a clear notice when sound, image or video is hard to tell apart from reality | SYS-003 | On-screen notice at the start of each video | In progress | Design ticket |

See also `ai-data/ai-content-transparency-plan` for the labelling design.

## 4. High-performance AI safety (Article 32)

<!-- Guidance: Applies when cumulative training compute is at or above the level set in the decree. Most utilisation business operators will record "not triggered", but keep the reasoning. -->

| Question | Answer |
|---|---|
| Training compute of models we develop | *Example:* 4.0 × 10^23 FLOP for SYS-002 |
| At or above the decree level? | *Example:* No, 250 times below 10^26 FLOP |
| If yes: lifecycle risk identification and mitigation, safety incident monitoring and response, report of results to MSIT | [Plan reference] |

## 5. High-impact AI confirmation (Article 33)

<!-- Guidance: Before offering a system, check whether it falls in a high-impact area listed in Article 2 (for example energy, drinking water, health care and medical devices, nuclear facilities, biometric analysis in criminal investigations, decisions on employment or loans, transport, public services, and student assessment in early childhood to secondary education). You may ask MSIT for confirmation. -->

| System | Area considered | Significant effect on life, safety or fundamental rights? | Conclusion | MSIT confirmation requested? |
|---|---|---|---|---|
| SYS-002 *(Example)* | Student assessment in secondary education | Scores feed teacher grading decisions | Treated as high-impact pending confirmation | Yes, YYYY-MM-DD |
| SYS-001 *(Example)* | None | Practice tutoring, no decision about the learner | Not high-impact | No |

## 6. High-impact AI responsibilities (Articles 34 and 35)

<!-- Guidance: For each high-impact system, record each measure from Article 34 and the impact assessment effort under Article 35. Keep the documents, because the Act expects them to be kept. -->

| ID | Measure (summary) | Article | System | Evidence | Status |
|---|---|---|---|---|---|
| KR-H-001 *(Example)* | Risk management plan in place and operated | 34 | SYS-002 | `ai-data/ai-risk-assessment` RA-07 | Done |
| KR-H-002 *(Example)* | Explanation of results, main criteria and an outline of training data, to the extent technically possible | 34 | SYS-002 | Teacher-facing explanation page | In progress |
| KR-H-003 *(Example)* | User protection measures | 34 | SYS-002 | Appeal route for students and parents | Done |
| KR-H-004 *(Example)* | Human management and supervision | 34 | SYS-002 | `ai-data/human-oversight-design` HOD-2 | Done |
| KR-H-005 *(Example)* | Documents showing the safety and trust measures | 34 | SYS-002 | This assessment and linked records | Done |
| KR-H-006 *(Example)* | Effort to assess the impact on fundamental rights before use | 35 | SYS-002 | `ai-data/ai-impact-assessment-42005` AIIA-03 | Planned |

## 7. Domestic representative (Article 36)

<!-- Guidance: A business operator without an address or place of business in Korea that meets a decree threshold designates a representative in Korea and reports it to MSIT. The operator remains responsible for the representative's acts. -->

| Threshold (see section 2) | Our figure for the prior year | Met? |
|---|---|---|
| Total revenue | *Example:* 180 billion KRW | No |
| Revenue from AI services | *Example:* 12 billion KRW | Yes |
| Average daily Korean users | *Example:* 140,000 | No |
| **Representative required?** | Any one threshold met | *Example:* Yes |

Representative: [name, address in Korea, date reported to MSIT].

## 8. Enforcement and action plan

<!-- Guidance: Record exposure and the actions that close open items. MSIT can investigate (Article 40) and issue corrective or suspension orders. Fines under Article 43 cover listed breaches such as failing to give advance notice, failing to designate a representative or failing to follow an order. -->

| ID | Action | Closes | Owner | Due | Status |
|---|---|---|---|---|---|
| ACT-001 *(Example)* | Add realistic-media notice to avatar videos | KR-T-003 | [@product-lead] | YYYY-MM-DD | Open |
| ACT-002 *(Example)* | Publish explanation page for essay scores | KR-H-002 | [@ml-lead] | YYYY-MM-DD | Open |
| ACT-003 *(Example)* | Complete impact assessment for SYS-002 | KR-H-006 | [@impact-lead] | YYYY-MM-DD | Open |
| ACT-004 *(Example)* | Designate and report domestic representative | Section 7 | [@legal] | YYYY-MM-DD | Open |

Track the Korean entries alongside other jurisdictions in
`compliance/ai-regulatory-applicability-register`.

## 9. References

<!-- Guidance: Use the Korean text as the binding version; translations are for reference only. -->

- Framework Act on the Development of Artificial Intelligence and the
  Establishment of a Foundation for Trust, Act No. 20676 (2025), as
  amended by Act No. 21311 (2026): Korean text at https://www.law.go.kr
  and English translation by the Korea Legislation Research Institute at
  https://elaw.klri.re.kr/eng_service/lawViewTitle.do?hseq=73499
- Enforcement Decree of the Act and MSIT guidelines on transparency,
  high-impact AI confirmation, high-performance AI safety, operator
  responsibilities and impact assessment (Korean, on MSIT and law.go.kr).
- Related templates: AI regulatory applicability register
  (`compliance/ai-regulatory-applicability-register`), AI content
  transparency plan (`ai-data/ai-content-transparency-plan`), AI risk
  assessment (`ai-data/ai-risk-assessment`), NIST AI RMF profile
  (`ai-data/nist-ai-rmf-profile`).

## Standard and licence note

Based on the Framework Act on the Development of Artificial
Intelligence and the Establishment of a Foundation for Trust of the
Republic of Korea, as amended to Act No. 21311 (20 January 2026), with
article numbers taken from the KLRI English translation, and on public
summaries of the Enforcement Decree as checked in October 2026. All
duties are paraphrased in original wording; no statutory text is
reproduced. The organisation and figures are fictional. Check the
current official text, the decree and MSIT guidelines, and consult a
qualified Korean professional. This is a documentation format, not
legal advice. Template licence: MIT.