Artificial Intelligence(AI) is a technology that implements human cognitive abilities in computers, and it has established itself as a core driving force of the Fourth Industrial Revolution alongside the digital transformation accelerated by the COVID...
Artificial Intelligence(AI) is a technology that implements human cognitive abilities in computers, and it has established itself as a core driving force of the Fourth Industrial Revolution alongside the digital transformation accelerated by the COVID-19 pandemic. In particular, recently emerged Generative AI has brought innovation to the creation of knowledge information and problem-solving methods, dramatically improving work productivity and generating immense economic value across industries, including specialized fields. However, behind these technological advancements, serious side effects and threats to national security persist, such as the dissemination of false information using deepfakes, the erosion of social trust, the deepening of information bias, and the potential for AI to degenerate into a tool for sophisticated cyber attacks and the development of biochemical weapons. This study stems from the recognition that, amidst these circumstances, it is urgent to establish legal and institutional regulations for the sustainable development of AI technology and the establishment of 'Trustworthy AI'. In particular, the purpose is to analyze regulatory trends in major countries around the world and critically review Republic of Korea's AI-related laws, which came into full effect in January 2026, in order to derive concrete legal improvement measures that can harmonize technological innovation with the protection of human rights. Looking at the trends in AI legislation enactment among leading global technology nations, the European Union(EU) established a four-tiered regulatory framework based on risk levels through the world's first 「Artificial Intelligence Act」. This framework is characterized by prohibiting, in principle, technologies with the potential to infringe on human rights, such as real-time biometric recognition, while imposing strict obligations on high-risk AI and levying severe punitive fines for violations. In contrast, the United States has moved away from the regulation-centric stance of the Biden administration, which sought to balance AI technological innovation and risk. With the recent announcement of the 「AI Action Plan」 under the second Trump administration, the United States has abandoned this regulation-centric approach and is rapidly shifting toward a strategy centered on innovation and competitiveness to secure a technological advantage in its hegemonic competition with China. China is also implementing a tight, grid-like regulatory framework focused on preventing social unrest and threats to the regime. It is rapidly completing its regulatory system by mandating real-name verification for deepfake service users and stipulating that AI-generated content must align with core socialist values. Amidst these global trends, Republic of Korea has pursued the enactment of the AI Framework Act, demonstrating a comprehensive and flexible legislative attitude that emphasizes industrial development. However, while the current legal framework is significant in that it has established a legal foundation, it faces challenges such as a lack of specific criteria for balancing promotion and regulation, and reduced legal stability and predictability as key regulatory matters are deferred to enforcement decrees or guidelines. In particular, regarding the guarantee of human rights in high-risk areas and the prevention of deepfake misuse, the lack of regulations regarding General-Purpose AI may lead to confusion within the industry and social distrust. Therefore, this study proposes two approaches—a complete revision and a partial revision—for the amendment of the Korean AI Framework Act, aiming to harmoniously realize the two values of AI technology development and ensuring reliability. Regarding the complete revision, first, the definitions of AI and AI systems must be advanced to align with international standards, such as those of the OECD and the European Union, and the roles of service providers must be clearly reclassified into providers and operators to clarify the locus of liability. Second, to protect human rights and safety, it is necessary to introduce a differential regulatory framework based on the risk level of AI systems and to codify the principle of prohibiting AI that undermines human dignity. Third, the existing concept of high-impact AI should be expanded to include public and sensitive areas affecting citizens' fundamental rights, such as investigation, trials, and employment, in addition to health and medical care, to enable effective risk management. Fourth, the scope of General-Purpose AI must be specified to ensure the effectiveness of technical safety assessments and copyright compliance, and the scope of exceptions to the law for specific purposes, such as national defense and research, must be reasonably adjusted to prevent research innovation from being stifled. Fifth, Notification methods for Generative AI content must be standardized at the national level to guarantee users' right to awareness, and the rights of users and affected parties (such as the right to demand explanations and the right to raise objections) must be legally stipulated. Sixth, recognizing that AI can become an ideological weapon if trained with data biased by specific forces from a national security perspective, the introduction of a government certification system for sectors directly related to national security should be considered to safeguard the values of liberal democracy. Finally, to ensure effective regulatory compliance, penalty provisions, such as punitive fines at the level of the European Union, should be strengthened; however, on-site inspections should be conducted supplementarily to ensure procedural legitimacy and alleviate the burden on the industry. As part of the partial amendment proposal, specific supplementary measures are presented to address the immediate risks posed by AI. Specific standards regarding deepfakes must be established within the definition of AI ethics and ethical principles, and relevant response strategies must be included when formulating the AI Basic Plan. To prevent deepfake-related crimes, the state's responsibilities for prevention and users' right to demand explanations must be strengthened, and the development of detection technologies and the overhaul of the damage compensation system must be carried out under the leadership of the National Artificial Intelligence Strategy Committee. Furthermore, to prevent the exploitation of cyber attacks, it is necessary to codify relevant concepts into law and mandate the Security by Design principle—which eliminates security vulnerabilities from the development stage—as well as Red Team testing. In terms of data, we must move beyond mere quantitative expansion to ensure public accountability through measures such as data card disclosure and third-party verification, which secure data diversity and prevent bias. Moreover, to serve as a safeguard for the public, we must prevent the mass production of so-called Black Box AI by mandating the retention of core logs and data documentation detailing algorithm operating principles. In conclusion, the Korean AI Framework Act must move beyond a perspective centered on self-regulation and industrial protection to stipulate a minimum framework capable of guaranteeing human safety and accountability. A balanced regulatory system is required that prioritizes human dignity and constitutional values while ensuring corporate flexibility through harmonization with global norms, thereby protecting national security and the rights and interests of citizens. This study holds academic and policy significance in that it examines new types and cases of risks caused by AI in a rapidly changing technological environment and proactively suggests directions for improving the Korean AI Framework Act by reflecting recent regulatory trends in leading technology countries. If follow-up research is conducted to address derivative issues such as AI governance, copyright, and personal information, more practical response measures could be devised.