Regulations Risk AI Framework Act’s Strengths and Global Best Practice Potential; New Report Calls for Amendments

SEOUL-By uniting strategy, promotion, and regulation in a single law, South Korea's AI Framework Act serves as a powerful policy instrument to guide the country's artificial intelligence ambitions. But the Act's blunt regulatory provisions threaten to undermine its strong foundations in strategy and industrial policy, jeopardizing South Korea's long-term global competitiveness, according to a new report from the Information Technology and Innovation Foundation's Center for Korean Innovation and Competitiveness.

The Act requires substantial amendments and must be carefully implemented, according to ITIF's report. Otherwise, its regulatory flaws-such as overly broad definitions, rigid R&D mandates, SME-first rules, invalid compute thresholds, and transparency labeling requirements-could outweigh its strengths.

"This law is the first in the world to integrate AI strategy, industrial promotion, and regulation in a single statute, so it has the potential to serve as a best-practice model for other countries. But its single-instrument design also means that flaws in one policy lever will inevitably weaken the others," said Sejin Kim, associate director of ITIF's Center for Korean Innovation and Competitiveness (Seoul). "While the Act establishes a coordinated national AI strategy and solid industrial policies, by prioritizing procedures over performance, it undermines the scale and flexibility South Korea needs to secure a lasting edge in the global AI race."

With the Act set to take effect in January 2026 and a draft Enforcement Decree now under review by South Korea's Ministry of Science and ICT (MSIT), the report analyzes the strengths and weaknesses of the law, including its overall structure, specific provisions, and draft enforcement ordinance.

ITIF calls for structural amendments by the National Assembly and balanced, practical implementation through MSIT's final Enforcement Decree.

• For the National Assembly: Narrow the definition of AI, eliminate compute thresholds, redesign high-impact AI rules to focus on performance, make R&D mandates flexible, remove SME-first language, replace mandatory labeling with voluntary standards, and scale penalties proportionally with a grace period before fines.
• For MSIT's Enforcement Decrees: Clarify that support is size-neutral, align oversight with sectoral expertise, safeguard confidentiality in data requests, and ensure standards are industry-led, with the government as the convenor.

Among its extensive list of proposed changes, the report highlights three key priorities:

1. Delete the compute threshold (Article 32). Computing use is not a reliable proxy for risk. Oversight should shift to post-deployment, performance-based evaluation (Article 12, strengthening AISI authority).
2. Introduce a grace period before fines (Article 43). Companies need warnings and time to adapt before penalties take effect, followed by strong enforcement.
3. Redesign high-impact AI standards (Articles 33-35). Replace documentation-heavy obligations with measurable performance criteria, coordinated through KRISS and sectoral ministries.

"The Act's regulatory provisions undercut its strong strategy and promotion agenda, and risk overshadowing the progress it could deliver," said Hodan Omaar, senior manager for AI policy at ITIF's Center for Data Innovation (Washington). "If policymakers refine how the Act defines risk, scopes obligations, and designs remedies, oversight will better target novel and consequential AI challenges without sweeping in ordinary tools. That will not only address harmful risks and provide stronger protections for Korean citizens, but also create the conditions for innovation to thrive."

As the first of its kind, Korea's integrated model could be emulated abroad, creating a "Seoul Effect." If regulatory flaws remain, Korea risks exporting restrictive rules that hurt domestic innovators and regional competitiveness. Conversely, if Korea makes structural fixes now and adopts balanced implementation, the Act could become a global best practice for coordinated AI policy.

"The AI Framework Act will set the course for South Korea's AI trajectory for the next decade. One-size-fits-all rules threaten to blunt those gains," said ITIF President Robert D. Atkinson (Washington). "The way forward is decisive and practical: tighten the statute to fix structural flaws and use the final Enforcement Decrees to implement balanced, risk-based, performance-focused rules. If Korea does this, the Act will catalyze innovation that improves daily life and strengthen its industrial competitiveness."

Read the report.

Contact: Sydney Mack, [email protected]