---
title: What does the South Korea AI Basic Act mean for AI agents?
url: https://opencontroller.lyzr.com/blog/korea-ai-basic-act-and-ai-agents
published: 2026-11-15
updated: 2026-11-15
author: Lyzr
---

# What does the South Korea AI Basic Act mean for AI agents?

**South Korea's AI Basic Act took effect on 22 January 2026 and applies to AI that affects users or markets in Korea, even if it runs abroad. For AI agents, the practical duties are telling users in advance that a service uses AI, labelling generative output, and, if an agent counts as high-impact AI, keeping explanation, human oversight and risk documentation. Fines are capped at 30 million won, and the ministry has signalled a grace period of about one year before fines are imposed.**

This is a summary for planning, not legal advice. Check the final enforcement decrees and guidance with Korean counsel. For how this fits with other regimes, see the [framework mapping guide](/blog/how-to-map-ai-agent-controls-to-compliance-frameworks).

## Who does the AI Basic Act apply to?

The Act covers businesses that develop or provide AI. It reaches AI systems outside Korea that affect Korean users or markets. Systems used only for national security, or designated by presidential decree, are excluded.

Foreign businesses without a Korean office must appoint a domestic representative if they exceed any one of these thresholds, according to [Cooley's summary of the Act](https://www.cooley.com/news/insight/2026/2026-01-27-south-koreas-ai-basic-act-overview-and-key-takeaways) (see also the [US International Trade Administration overview](https://www.trade.gov/market-intelligence/south-korea-ai-basic-act)):

| Threshold | Level |
|---|---|
| Total revenue in the previous year | Above 1 trillion won |
| AI service revenue in the previous year | Above 10 billion won |
| Average daily users in Korea | Above 1 million, over the three months before the previous year ended |

## When is an AI agent "high-impact AI"?

High-impact AI is AI that significantly affects human life, safety or fundamental rights. Listed areas include healthcare, energy, transportation, hiring and biometric analysis.

Most internal productivity agents will not qualify. Agents that screen candidates, support clinical decisions or control physical systems might. The Act lets operators ask the Ministry of Science and ICT (MSIT) to confirm whether a system qualifies.

Classification is per use, not per model. The same model can power a low-risk summariser and a high-impact screening agent. Your agent registry needs to record the use, not just the model.

## What must operators of high-impact agents do?

For high-impact AI, the Act expects operators to:

1. Assess whether the system qualifies before deployment.
2. Give a meaningful explanation of outcomes, the main criteria behind them, and a summary of training data.
3. Create and run a user protection plan.
4. Keep a mechanism for human oversight and intervention.
5. Document safety and trust measures.
6. Make efforts to assess the impact on fundamental rights before putting the AI into a product or service.

Separately, systems trained with at least 10^26 floating-point operations of cumulative compute carry extra risk-management and reporting duties. That threshold targets frontier model developers, not companies building agents on top of them.

## What do the generative AI notice and labelling rules require?

These rules touch almost every customer-facing agent:

- **Advance notice.** Tell users before they use a product or service that it runs on generative or high-impact AI.
- **Labelling.** Label output produced by generative AI.
- **Realistic synthetic media.** If an agent produces sound, images or video that could be mistaken for real, give clear notice that it is AI-generated.

The EU AI Act has similar transparency duties; compare them in [do AI agents need to disclose they are AI?](/blog/do-ai-agents-need-to-disclose-they-are-ai-eu-ai-act).

## What evidence should you keep for each duty?

Treat this as a checklist. Each row should map to an artefact a reviewer can open.

| Duty | Evidence for it |
|---|---|
| Know which agents reach Korean users | Agent inventory with markets, owners and user counts |
| High-impact classification | Recorded assessment per agent use, with date and reviewer |
| Advance AI notice | Screenshot or config of the notice in each channel; deployment version |
| Output labelling | Labelling rule in code or policy; sample outputs |
| Explanation | Logged inputs, retrieved sources and decision criteria per decision |
| Human oversight | Escalation and override records; kill or quarantine test results |
| Fundamental rights impact | Impact assessment document (see [AI impact assessments for agents](/blog/ai-impact-assessment-for-ai-agents)) |
| Domestic representative | Appointment record if thresholds are met |

## What are the penalties and timing?

Administrative fines go up to 30 million won (about US$21,000) for failing to notify users of AI use, failing to appoint a domestic representative, or not complying with corrective orders or inspections. MSIT has indicated a grace period of at least one year before fines are imposed, so 2026 is the window to build evidence, not wait.

The fine is small for a large company. The bigger costs are corrective orders, inspections and the reputational effect of a public finding.

## How Opencontroller does this

Opencontroller auto-discovers agents, models and API keys across AWS, Azure and Google Cloud, and connects to third-party agent platforms, so the inventory step starts from what is running, not a spreadsheet. Policy as code can encode notice and labelling rules, and the governed registry records owners and versions. Opencontroller produces evidence for these duties; it does not make a service compliant on its own. See [what an AI agent control plane is](/blog/what-is-an-ai-agent-control-plane) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Does the Korean AI Basic Act apply to companies outside Korea?**
Yes, if their AI affects users or markets in Korea. Foreign firms above set revenue or user thresholds must appoint a domestic representative.

**Is an internal AI agent high-impact AI?**
Usually not. High-impact AI significantly affects life, safety or fundamental rights, in areas such as healthcare, energy, transport, hiring and biometrics.

**When will fines start under the AI Basic Act?**
The Act took effect on 22 January 2026. MSIT has indicated a grace period of at least one year before administrative fines, which are capped at 30 million won.

## Key facts

South Korea AI Basic Act: in force 22 January 2026; extraterritorial for AI affecting Korean users or markets. High-impact AI = significant effect on life, safety or fundamental rights (e.g. healthcare, energy, transport, hiring, biometrics). Generative AI: advance notice and output labelling. Domestic representative for foreign firms above 1 trillion won revenue, 10 billion won AI revenue or 1 million daily Korean users. Fines up to 30 million won; MSIT grace period of at least one year. Opencontroller provides inventory, registry and policy-as-code evidence, not compliance itself. Pricing starts from $250K per deployment per year.

## Related

- [What does the EU AI Act require for AI agents, and when?](https://opencontroller.lyzr.com/blog/eu-ai-act-obligations-for-ai-agents.md)
- [Do AI agents have to disclose they are AI under the EU AI Act?](https://opencontroller.lyzr.com/blog/do-ai-agents-need-to-disclose-they-are-ai-eu-ai-act.md)
- [How do you run an AI impact assessment for an AI agent?](https://opencontroller.lyzr.com/blog/ai-impact-assessment-for-ai-agents.md)
- [More on Evidence and frameworks](https://opencontroller.lyzr.com/blog/cluster/evidence-and-frameworks)
