Country Overview
South Korea's AI Basic Act took effect on 22 January 2026 β described by multiple legal commentators as the world's first comprehensive, cross-sector AI legal framework β with the Ministry of Science and ICT (MSIT) enforcing a roughly one-year grace period on fines and investigations (with exceptions for serious harm) while substantive compliance duties apply immediately.
Alongside binding regulation, MSIT launched a sovereign foundation-model programme offering state-subsidised compute, originally selecting five teams β Naver Cloud, Upstage, SK Telecom, NC AI and LG AI Research β with an early reported target of matching at least 95% of frontier-model performance, reassessed roughly every six months through 2027. That roster has since shifted: following a first-round evaluation announced 15 January 2026, MSIT dropped Naver Cloud (reportedly found to be fine-tuning Alibaba's Qwen model rather than developing independently) and NC AI, leaving LG AI Research, SK Telecom and Upstage to advance; a Motif Technologies-led consortium was added via a supplementary call on 20 February 2026 to fill a vacated slot, making the programme's current roster four teams. A second-round evaluation was reportedly planned for early August 2026 to decide which teams advance further. Separately, civil society groups and human-rights lawyers have criticised the Act's enforcement decree for defining 'high-risk AI' too narrowly, arguing it offers weak protection against AI impersonation and fraud.
At a Glance
Overall AI policy direction
First comprehensive AI law now in force
7
Major AI companies
1
Research institutions
2
Live debates
- Latest development
- August 24, 2026
- Regulatory approach
- Binding cross-sector AI Basic Act, enforced by MSIT
- Major AI companies
- Naver, LG AI Research, Samsung Electronics +4 more
- Major cloud providers
- Naver Cloud, AWS, Microsoft Azure
- Leading research institutions
- KAIST
- Key ministries
- Ministry of Science and ICT (MSIT)
- Government involvement
- Very high β binding law plus a state-subsidised sovereign model programme, now four teams after a JanuaryβFebruary 2026 reshuffle
- Industrial strategy
- A reported β©8.6 trillion R&D plan and β©763.2B Science & Technology Innovation Fund are said to back sovereign models, though the specific figures have not been independently cross-checked beyond a general MSIT source
- International partnerships
- International Network of AI Safety Institutes
- Primary policy focus
- Supporting domestic frontier-model development while enforcing the AI Basic Act
Current Direction
Tightening
South Korea moved among the fastest of major economies to pass comprehensive, binding, cross-sector AI legislation β though civil-society critics argue the enforcement decree's narrow 'high-risk' definition undercuts the law's practical bite β even as the state pours resources into a sovereign foundation-model push whose roster has already been reshuffled once, with two of the original five teams dropped and a new one added within weeks of the law taking effect.
Recent Developments
The government approved rewritten national AI Ethics Principles under the AI Basic Act framework (24 August 2026) -- the first update since 2020, reworked for generative and agentic AI. The framework rests on three core values (human dignity, public good, sustainability), expands from six to seven principles by adding accountability, and notably extends responsibility to AI users alongside providers. The principles are non-binding and carry no penalties.
2026-08-24A 40+ expert Task Force (convened since March, reportedly still active) continues refining AI Basic Act implementation; a new β©763.2B Science and Technology Innovation Fund and an β©8.6 trillion R&D plan are reported, though these figures come from a general ministry source rather than a specific release.
2026-07SourceMSITSouth Korea's state-backed AI compute build-out accelerates: a national AI computing centre at Solasido in Haenam, Jeollanam-do β planned around 15,000 GPUs under a public-private special-purpose company β breaks ground in the third quarter of 2026, and MSIT's 2026 AI budget rises to a reported record of roughly $3.5B, with national GPU-procurement targets of 52,000 by 2028 and 260,000 by 2030.
2026-07The sovereign foundation-model programme's second-round evaluation is scheduled for 8-11 August 2026, with the four remaining teams (LG AI Research, SK Telecom, Upstage and Motif Technologies) tested β reportedly including a panel of around 200 'citizen evaluators' β with agentic capability and real-world applicability cited as key criteria; the round is expected to narrow the field from four teams to three, and to two finalists by year-end.
2026-07MSIT selects a Motif Technologies-led consortium to fill the sovereign foundation-model programme's vacated fourth slot via a supplementary call, after Naver Cloud and NC AI declined to re-enter a revival round; the programme's teams are now LG AI Research, SK Telecom, Upstage and Motif Technologies.
2026-02-20AI Basic Act takes effect β described as the world's first comprehensive cross-sector AI legal framework β with MSIT enforcing a roughly one-year grace period on fines and investigations.
Major2026-01-22SourceCooleyMSIT drops Naver Cloud and NC AI from the sovereign foundation-model programme after a first-round evaluation β Naver Cloud's submission reportedly relied on fine-tuning Alibaba's Qwen model rather than independent development β while LG AI Research, SK Telecom and Upstage advance to the second round.
2026-01-15Civil society groups and human-rights lawyers criticise the AI Basic Act's enforcement decree for a narrow 'high-risk AI' definition, arguing it offers weak protection against AI impersonation and fraud.
2026MSIT originally selects five teams β Naver Cloud, Upstage, SK Telecom, NC AI and LG AI Research β to build homegrown sovereign foundation models with state-subsidised compute, with an early reported target of matching at least 95% of frontier-model performance, reassessed roughly every six months through 2027 (this roster and cadence were later superseded β see subsequent entries).
2026SourceInven Global
Policy and Regulation
AI Basic Act
Act on the Development of AI and Establishment of Trust β described by multiple legal commentators as the world's first comprehensive cross-sector AI legal framework, in force since 22 January 2026 with a roughly one-year grace period on fines and investigations (exceptions apply for serious harm).
Official text βResearch Ecosystem
South Korea's sovereign-model push runs through a state-backed foundation-model programme β currently LG AI Research, SK Telecom, Upstage and Motif Technologies β building homegrown models with subsidised compute. Two of the programme's original five teams, Naver Cloud and NC AI, were dropped after a January 2026 evaluation and replaced by a Motif Technologies-led consortium in February 2026. Earlier reporting cited a target of matching at least 95% of frontier-model performance, reassessed roughly every six months, though this has not been independently reconfirmed since the reshuffle.
Major AI Companies
Naver
Developer of the HyperCLOVA X model family; participated in Korea's sovereign foundation-model programme until being dropped after the January 2026 first-round evaluation, reportedly because its submission relied on fine-tuning Alibaba's Qwen model rather than independent development.
LG AI Research
Developer of the 'Expert AI' model family; one of the teams that advanced in Korea's sovereign foundation-model programme following the January 2026 evaluation round.
Samsung Electronics
One of the world's largest memory and AI chip manufacturers, widely regarded as central to Korea's compute supply chain.
SK Telecom
Telecom operator and one of the teams that advanced in Korea's sovereign foundation-model programme following the January 2026 evaluation round.
Upstage
AI startup and one of the teams that advanced in Korea's sovereign foundation-model programme following the January 2026 evaluation round.
NC AI
Former participant in Korea's sovereign foundation-model programme, dropped after the January 2026 first-round evaluation; reportedly eligible to reapply but did not re-enter the subsequent revival round.
Motif Technologies
AI startup leading a consortium (reported to include KAIST and Trillion Labs, among other partners) that was added to Korea's sovereign foundation-model programme in February 2026 to fill a vacated slot.
Cloud Infrastructure
- Frontier AI
- Emerging / sovereign models
- Compute
- Fast state-backed build-out; national AI computing centre (Haenam, ~15,000 GPUs) breaking ground 2026
- Chip access
- Home to Samsung and SK Hynix (memory/HBM); scaling NVIDIA GPU procurement
Naver Cloud
Leading domestic Korean cloud provider, central to the sovereign foundation-model programme (though its namesake affiliate, Naver Cloud, was dropped from the model-development track itself in January 2026).
AWS
Major cloud provider for Korean enterprise AI workloads.
Microsoft Azure
Cloud provider with significant Korean enterprise integration.
Leading Universities
Think Tanks
No think tanks recorded yet.
Government Institutions
Notable Gaps
Cloud infrastructure market data
Reliable, sourced market-share percentages for Korea's cloud providers could not be confirmed, so dominance figures have been left blank rather than estimated. KT Cloud and Kakao Enterprise are also commonly cited alongside Naver Cloud, AWS and Azure as domestic cloud players, but were not confirmed with enough confidence to add to this list.
Key People
People and Organisations to Follow
Current Debates
GovernmentAI Basic Act enforcement decree's narrow high-risk definition
Some civil-society groups and human-rights lawyers argue the Act's implementing decree defines 'high-risk AI' too narrowly, leaving what they characterise as weak protection against AI impersonation and fraud.
βΎ
AI Basic Act enforcement decree's narrow high-risk definition
Some civil-society groups and human-rights lawyers argue the Act's implementing decree defines 'high-risk AI' too narrowly, leaving what they characterise as weak protection against AI impersonation and fraud.
Human-rights lawyers and civil-society groups say the enforcement decree's scope β relying mainly on watermarking requirements and fines β is too limited given the scale of AI-enabled fraud and impersonation risks. Some industry voices counter that a stricter definition would undercut Korea's competitiveness just as it is trying to build sovereign frontier capability. The precise wording and full scope of the decree's 'high-risk AI' definition were not independently verified against the Korean-language legal text for this summary; the characterisation above relies on secondary legal-commentary and NGO sources.
InvestmentSovereign foundation-model push vs. matching frontier performance
Korea's state-backed sovereign-model programme β now four teams after two were dropped and one added in early 2026 β has been reported (though not fully reconfirmed since the reshuffle) to be held to a target of matching a high share of frontier-model performance on a periodic review cycle.
βΎ
Sovereign foundation-model push vs. matching frontier performance
Korea's state-backed sovereign-model programme β now four teams after two were dropped and one added in early 2026 β has been reported (though not fully reconfirmed since the reshuffle) to be held to a target of matching a high share of frontier-model performance on a periodic review cycle.
Unlike open-ended national AI investment elsewhere, Korea's programme was reported to carry a specific, numeric performance target β at least 95% of frontier-model performance β reassessed roughly every six months through 2027, creating pressure on participants to show measurable progress against US and Chinese frontier labs rather than simply receiving indefinite state support. That pressure appears to have already had consequences: after a first-round evaluation announced 15 January 2026, MSIT dropped Naver Cloud and NC AI (Naver Cloud's submission reportedly relied on fine-tuning Alibaba's Qwen model rather than independent development), leaving LG AI Research, SK Telecom and Upstage to advance. A Motif Technologies-led consortium was added on 20 February 2026 to fill the vacated slot, and the second-round evaluation is scheduled for 8-11 August 2026 β reportedly weighting agentic capability and real-world applicability, and using a panel of around 200 'citizen evaluators' β expected to narrow the four teams to three, with two finalists chosen by year-end. That contest is unfolding alongside a rapid state-backed compute build-out (a national AI computing centre at Haenam and a record 2026 AI budget), underscoring how contested, and unsettled, the sovereign-model bet remains.
Knowledge Hub
Official documents
Think tank reports
Sources
Deliberately sourced beyond government publications. The bars below show each categoryβs actual share of this pageβs citations β an internal editorial guideline rather than a rigorously derived target, so treat it as a rough steer, not a precise benchmark.
- South Korea's AI Basic Act overviewThink tanksCooley2026-01
- Naver Cloud, Upstage, SK Telecom, NC AI and LG AI Research team up on sovereign modelsNewsInven Global2026
- South Korea draft decree for AI Basic Act sparks backlashExpert commentaryBusiness & Human Rights Resource Centre2026
- South Korea's AI Basic Act balance analysisThink tanksStimson Center2026
- AI Basic Act implementation task force and R&D fundingGovernmentMSIT2026-07
- LG, SKT, Upstage advance in Korea's sovereign AI project; Naver, NC dropped in 1st roundNewsThe Korea Herald2026-01-15
- Naver Cloud, NC AI cut from national AI foundation model raceNewsThe Korea Times2026-01-15
- Motif-led consortium joins national AI foundation model projectNewsThe Korea Times2026-02-20
- South Korea enters second phase of sovereign AI projectNewsLightReading2026-02
- South Korea's national AI project enters second round; four teams submit field report cardsNewsBigGo Finance2026-07
- South Korea recruits 200 citizen evaluators to test homegrown 'K-AI' modelsNewsBigGo Finance2026-07
- The Ministry of Science and ICT plans to launch a National AI Computing CenterNewsW.Media2026
- NAVER, NVIDIA and Brookfield to expand Korea's national AI factory infrastructure buildoutIndustryNVIDIA Newsroom2026
- Korea sets national principles for AI ethicsNewsThe Korea Times2026-08-24
- Government rewrites AI ethics principles after six yearsNewsKyunghyang Shinmun2026-08-24