[비즈한국] The 'AI Framework Act', which addresses the development and safety of artificial intelligence (AI), went into full effect on the 22nd. This legislation aims to enhance domestic AI competitiveness while establishing a foundation of trust in AI utilization by creating safeguards, such as mandatory watermarking for AI-generated content. With generative AI becoming integrated into daily life, a legal framework covering the entire industry is now in operation. However, there is room for interpretation regarding the specific criteria for 'high-impact AI,' which is the core of the regulation, and the practical application of transparency requirements is still being discussed. Critics note that while the system is in place, the operational standards and verification frameworks remain vague, leaving much work to be done.
Distinguishing Real from Fake… AI Content Labeling Made Mandatory
A major change that users will notice with the implementation of the AI Framework Act is the application of 'transparency obligations.' AI service providers must ensure that users can clearly identify when AI was used to create content, such as images, videos, or audio, when distributing or providing such content. This is known as the 'AI watermark.'
The government explains that it will keep the display methods flexible to accommodate technological feasibility. Depending on the characteristics of the content, both visible labeling and invisible methods, such as embedding code into data for machine readability, are permitted. Specifically, for deepfake results that raise social concerns and are difficult to distinguish from reality, users must be clearly notified or alerted to the fact that the content was AI-generated.
These transparency obligations apply only to businesses providing AI products or services. This includes overseas businesses that provide AI products and services to domestic users. The obligations apply to the business that ultimately delivers the AI service to the user. For example, if company B uses an AI model developed by company A via an API in its own service, the transparency obligation falls on company B. In cases where companies like OpenAI, Google, or Naver035420 develop models and also provide the services directly, they bear the transparency obligations.
Individual users who simply use AI tools to consume content or post on personal social media are excluded from these regulations. For watermark methods by service, please refer to the Ministry of Science and ICT's transparency guidelines.
10 Areas of 'High-Impact AI' Directly Linked to Life and Safety
Another key aspect of the law is the 'High-Impact AI' system. According to the enforcement decree, AI utilized in 10 areas directly linked to life, safety, and fundamental rights (energy, drinking water, healthcare, nuclear energy, criminal investigation, employment, loan screening, transportation, public services, and education) is classified as high-impact AI.
Specific examples include 'Level 4 or higher fully autonomous vehicles (highly automated systems that do not require human intervention).' In fields that significantly affect human life and rights, such as medicine, recruitment, and loan screening, businesses must implement strengthened measures to ensure reliability, such as establishing risk management systems and user protection plans.
Businesses must self-assess whether their AI qualifies as 'high-impact AI.' If an AI business is uncertain, it may request confirmation from the Ministry of Science and ICT. However, there is a prevailing view that there is significant room for interpretation regarding what level of technology or service falls under the regulation. Inside and outside the industry, opinions clash between those who argue that "it should only be considered high-impact AI when a human does not intervene in final decision-making, like in full autonomous driving" and those who believe "it should include all algorithms that pose a risk of infringing on fundamental rights."
Amid criticism that specific criteria for practical application are lacking, the government plans to focus on institutional stability rather than harsh penalties, considering potential confusion in the industry following the law's implementation. While the act allows for fines of up to 30 million KRW for violations, to minimize corporate confusion, enforcement of these regulations will be deferred for at least one year.
Additionally, the government decided to establish and operate an 'AI Framework Act Support Desk' to handle company inquiries regarding the law. Starting next month, a 'System Improvement Research Group' consisting of industry experts, civic groups, and academics will also be formed to identify areas for system improvement.
Government to Provide 'One-Year Grace Period'… Platforms and Telecom Industry Respond Proactively
Major domestic platforms and telecommunications companies are adjusting their internal policies and governance as the AI Framework Act emphasizes social responsibility for AI technology.
Kakao035720 has codified the 'duty to include and notify of AI-based services' in its terms of service, which take effect on February 4. Naver will also strengthen its AI-based content usage policy starting on the 7th of next month. Late last year, Naver introduced an 'AI-generated content labeling' feature on its user-generated content platforms, including blogs, cafes, and Clips.
The telecommunications industry is focusing on strengthening enterprise-wide governance systems. SK Telecom017670 is conducting a 'Good AI' campaign for its members and upgrading its 'AI Governance Portal' to check for risks from the planning stage. KT030200 is setting out to build an AI ecosystem that ensures safety and reliability through its dedicated organization, the 'Responsible AI Center (RAIC).'
The government plans to prevent reverse discrimination against domestic companies by ensuring the law applies to overseas AI businesses that affect the public. It has also left open the possibility of extending the grace period, considering international trends such as the application of EU regulations (in 2027).
Kim Kyung-man, head of the Artificial Intelligence Policy Bureau at the Ministry of Science and ICT, emphasized, "80 to 90 percent of the AI Framework Act is about industrial promotion. Initially, we will focus on supporting companies to adapt to the system rather than enforcing regulations," adding, "While EU regulations impose fines of 35 million EUR (approx. 67 billion KRW) or 7% of total revenue, whichever is higher, our domestic law caps fines at 30 million KRW. It is a minimal regulation that does not stifle innovation."