The AI Framework Act took effect on Jan. 22. It is the first time in the world that a basic law covering both AI promotion and regulation has taken effect.

The government says it will build a foundation of trust in AI and grow the industry over the long term. The industry, meanwhile, is watching provisions that could immediately affect service operations and product design. With the introduction of the concept of “high-impact AI” and an obligation to ensure safety, some expect changes to the industrial landscape to be unavoidable.

PROMOTION TO REVITALISE INDUSTRY...GOVERNMENT SEEKS TO SHAPE AI SECTOR

The AI Framework Act focuses on supporting the sound use of AI while preventing misuse. To build a safety-and-trust foundation, it sets out provisions on AI ethics, testing and certification, securing transparency and safety, and high-impact AI. The enforcement decree spells out details, scope and methods for implementing safety-and-trust-related requirements, clarifying matters set out in the law.

The Ministry of Science and ICT said it reflected broad measures to promote the AI industry while minimising obligations and sanctions for AI operators under the principle of minimum necessary regulation.

With the law taking effect, the National AI Strategy Committee, chaired by the president, was upgraded from an advisory body to a statutory committee. It will serve as a control tower coordinating and deliberating overall AI policy. Some say the legal basis has created a foundation for pursuing a national AI strategy consistently.

Another feature of the AI Framework Act is that it clarified its status as a higher-level law. Unless another law contains special provisions related to AI technology or industry, matters concerning AI will follow this act.

It also includes items and procedures to foster industrial development, including AI research and development, building training data, support for AI adoption and use, support for start-ups, promoting AI convergence, securing specialised talent, and support for building AI data centres. The enforcement decree supplemented the details.

INDUSTRY ANXIOUS OVER HIGH-IMPACT AI AND LABELLING OBLIGATIONS

The provisions the industry cites as having the biggest impact under the AI Framework Act are rules on high-impact AI and on labelling AI-generated outputs.

High-impact AI refers to AI used in areas that could significantly affect people’s lives or rights, including energy, healthcare, recruitment, transport and education. Businesses using high-impact AI must establish systems managed by humans and put safety measures in place.

Industry feels burdened at this point. If the scope of AI subject to regulation is not clear, it is difficult to calculate risks in advance. The government’s current position is that targets include vehicles at Level 4 or above, which are fully autonomous driving. The industry, however, believes the scope of high-impact AI regulation will also expand quickly as AI technology develops rapidly.

The obligation to label AI-generated outputs is also a point of concern for the industry. Under the AI Framework Act, businesses providing products and services using high-impact AI or generative AI must inform users in advance that they are operated based on AI, and outputs created with generative AI must be labelled.

If AI-generated outputs are provided only within a service environment, they can be labelled relatively flexibly through UI display or logos. If AI-generated outputs are taken outside, clearer labelling is required.

When downloading or sharing AI-generated text, images or videos, the act calls for either labelling in a “human-recognisable method” such as visible or audible watermarks, or applying a “machine-readable method” such as metadata to AI-generated outputs after providing text or voice guidance. Deepfake video content that is hard to distinguish from fact must also include markings identifiable to the human eye.

The Ministry of Science and ICT said watermarking AI-generated outputs is a minimum safety device to prevent side effects such as misuse of deepfakes. The industry, however, still has concerns. Costs and user experience can vary depending on what is labelled and how at the actual service operation stage.

The AI Framework Act also sets out an obligation to ensure safety. The scope of the safety obligation must meet all requirements, including cumulative compute used for training of 10 to the power of 26 floating point operations (FLOPs) or more, application of state-of-the-art technology, and a risk of broadly and seriously affecting people’s fundamental rights.

GOVERNMENT SAYS IT WILL DEFER REGULATION FOR AT LEAST 1 YEAR, BUT UNCERTAINTY PERSISTS

As concerns persisted that the AI Framework Act could undermine growth potential in the domestic AI industry, the government brought out a “regulatory grace period” card. It aims to ease regulatory shock and encourage a soft landing so the market can adapt.

The act allows administrative fines of up to 30 million won for violations, but the government decided to defer this for at least 1 year. It also plans to conduct fact-finding investigations only in extremely exceptional cases that cause serious social problems such as fatal accidents or human rights violations.

Even so, industry anxiety remains. Concerns have been raised that gaps could exist throughout the law as the still-immature AI industry is handled institutionally. The industry expects confusion is likely to intensify as real cases accumulate after the law takes effect. An AI industry official said, “There is still a lot of confusion about who becomes subject to regulation and in what cases,” adding, “We are in a situation where we have to keep studying the law.”

The Ministry of Science and ICT plans to resolve confusion on the ground through an “AI Framework Act support desk”. The support desk is an organisation composed of experts who participated in drafting subordinate statutes. The experts will review and analyse the AI Framework Act and provide detailed consulting to companies. Consultation details will be kept confidential, and anonymous consulting will also be provided.