A bill to amend the AI Framework Act (인공지능 발전과 신뢰 기반 조성 등에 관한 기본법 일부개정법률안) moved from committee referral to committee review on 2026-09-08. As drafted, it would require AI-business operators that develop or provide "AI agents" — systems that plan and act autonomously across multiple steps, using external systems or tools, without a person's individual instruction for each step — to build in specific controls before deployment. It is a pending bill, not current law; no effective date or penalty is specified in the draft as summarized here.

What the bill would require

For AI-business operators developing or providing AI agents (draft Art. 32-2, new):

  • Pre-set the scope and limits of delegated authority before the agent operates.
  • Assess and manage risks arising from linkage to external systems or tools.
  • Implement human management/supervision and technical measures to prevent the agent from operating beyond its authorized scope.
  • The government would be authorized to provide consulting and advisory support to help operators meet these duties.

Separately, the bill would add a duty to the government's trust-infrastructure policy (draft Art. 29) to fund education and outreach so the public can use AI ethically and responsibly, linked to the digital-competency activities under Article 14 of the Digital Inclusion Act. This part addresses government policy, not a direct operator obligation.

What this means for you

  • This is not yet in force. It is a bill under committee review (소관위 심사) as of 2026-09-08. No threshold, effective date, or penalty has been set — track it before treating any of the above as a compliance requirement.
  • Check whether you offer an "AI agent" feature. The draft's target is any AI service that autonomously plans and executes multi-step tasks via external systems or tools without step-by-step human instruction — not AI features generally. If you have such a feature, or are building one, this bill is worth watching closely.
  • If it advances, expect to need: a documented delegation-of-authority scope for each agentic feature, a risk assessment covering external tool/system integrations, and human oversight/technical guardrails preventing out-of-scope action. Start informal documentation now if you already operate agentic features in Korea, so you are not building this from scratch if the bill passes.
  • Re-check this regime periodically. Committee review can stall, advance, or be folded into a broader revision of the AI Framework Act; the obligations and their trigger conditions may change materially before enactment.

Source: No source URL was provided in the input for this alert.

Munteok provides regulatory information, not legal advice.