South Korea's AI Basic Act Imposes Mandatory Watermarking for Generative AI Content, Adding a New Jurisdiction to the Global AI Compliance Map
South Korea's AI Basic Act — a six-chapter, 43-article statute — entered into force on 22 January 2026, making South Korea one of the first countries to implement comprehensive AI legislation after the EU AI Act. A core requirement is that content generated by AI must carry watermarks or other provenance indicators identifying AI origin, applying across generative AI systems. The law distinguishes between "high-impact AI" (systems posing greater risks) and "general AI", imposing prior-notice obligations, inspection, and certification requirements on operators of high-impact systems. A parliamentary amendment proposed by Representative Hwang Jeong-a of the Democratic Party would defer the generative-AI labelling rules — including watermarking — for three years. Enforcement gaps remain unresolved: it is not yet settled whether watermarking obligations fall on model providers, platform operators, UI-level interfaces, or all three. For global platforms and law firms advising technology clients, the Act creates a third major AI regulatory regime alongside the EU AI Act and emerging US state-level rules, increasing the compliance burden for multi-jurisdictional operations.
Why this matters
Mandatory AI watermarking is technically and legally novel: it requires standards bodies, regulators, and courts to define what constitutes a valid provenance mark, how detection is verified, and who bears liability if a watermark is stripped or bypassed. For UK and EU law firms advising technology clients on cross-border AI compliance, South Korea's Act creates a third regulatory tier that must be mapped against the EU AI Act and domestic UK AI governance frameworks. The proposed three-year deferral introduces further uncertainty — clients must plan for both the current regime and a potentially materially amended one.
On the Ground
A trainee on an AI compliance matter would be drafting a regulatory impact assessment memo mapping the South Korean Act's watermarking obligations against the client's existing EU AI Act compliance programme, marking up data processing agreements to capture provenance obligations, and drafting vendor due diligence questionnaires for AI tool providers.
Interview prep
Soundbite
Three major AI regulatory regimes now operate simultaneously — firms advising global tech clients need multi-jurisdiction compliance stacks, not single-country playbooks.
Question you might get
“How does South Korea's AI Basic Act watermarking requirement compare to the EU AI Act's approach to AI-generated content transparency, and what compliance challenges arise for a UK tech company operating in both jurisdictions?”
Full answer
South Korea's AI Basic Act, in force since January 2026, requires AI-generated content to carry watermarks identifying AI origin and imposes inspection and certification on high-impact AI operators. For law firms, this creates a third major AI regulatory framework — alongside the EU AI Act and US state-level rules — that multi-jurisdictional technology clients must comply with simultaneously. The watermarking obligation is particularly significant because it requires technical standards that do not yet exist in agreed form, meaning compliance programmes must be built on incomplete regulatory guidance. A pending amendment to defer the rules by three years adds planning uncertainty — clients need legal advice on how to build adaptable compliance frameworks rather than fixed-point solutions.
Sources
My notes
saved