Acknowledgement

We are deeply grateful to all the interviewees who generously shared their time, expertise, and insights with us for this research.

We would especially like to thank Mr. Chihhao Yu of IORG and g0v for sharing his insights into Taiwan’s information environment, the role of civic tech communities such as g0v, and IORG’s independent research on FIMI.

We are also deeply grateful to Ms. Huian Ho, Head of International Affairs at the Taiwan FactCheck Center, for sharing her insights into TFC’s international projects, regional cooperation among fact-checking organizations, and the challenges of sustaining such collaboration.

We would like to appreciate Ohyeon Kweon, Chairman of the social cooperative Parti, for sharing his profound insights both on civic and public initiatives engaging in promoting democratic values in relation to AI in the Republic of Korea.

We would also like to express our sincere gratitude to Ms. Summer Chen, founder of FactLink, for sharing her insights into FactLink’s work on prebunking, disinformation playbooks, and capacity-building for journalists and educators.

We are also grateful to Mr. Tim Niven, Deputy CEO of Doublethink Lab, for sharing his insights into Taiwan’s information space, particularly the evolving tactics and narratives of PRC-linked information manipulation.

Overview

1. Overview

Japan, South Korea, and Taiwan have all adopted both soft and hard law frameworks for AI governance but differ in strategic focus and implementation. Japan has introduced new frameworks and revised existing institutional arrangements in a relatively rapid manner. Beyond domestic enactment such as the AI Act and the AI Basic Plan—which grounds governance in democratic values including human rights, rule of law, diversity, and human-centricity articulated in the Social Principles of Human-Centric AI—Japan actively shapes international norms through the Hiroshima AI Process as well. South Korea, driven by its AI G3 vision to catch up with the U.S. and China, prioritizes industry-friendly deregulation through regulatory sandboxes. In parallel, Korea’s democrataic key message has shifted from the Yoon adrninistration’s Liberal Democracy to the current Lee adrninistration’s AI Basic Society, which emphasizes universal AI literacy and AI-assisted social services. Taiwan presents its distinctive approach under pressure from the PRC, while embedding AI governance within democratic values. Its Artificial Intelligence Basic Act emphasizes human autonomy, privacy, fairness, and accountability, supported by robust civil society infrastructure including g0v alongside independent fact-checking organizations.

1. Japan

1.1. Institutional Response and International Commitment

Japan’s approach to AI governance is characterized by relatively rapid institutional responses combined with continuous engagement in international rule-making. Japan enacted the AI Act in early 2025 and adopted the AI Basic Plan in December 2025. Together with earlier policy initiatives, these documents aim to balance risk and innovation while enabling agile responses to technological change, with the strategic goal (Cabinet Office, Government of Japan 2026) of making Japan “the most AI-friendly country in the world.” Rather than establishing a comprehensive regulatory framework similar to the EU’s AI Act, Japan’s governance model relies on soft-law instruments. These include the Social Principles of Human-Centric AI (2019), the AI Business Guidelines (2024, revised in 2025), and the Guideline for Ensuring the Appropriateness of Research and Development and Utilization of AI-Related Technologies (2025). Reflecting the societal impact of AI, Japan has also revised, or is seeking to revise, existing legal frameworks in several areas, including, as discussed below, copyright, property rights, and privacy protection.

Domestic discussions on governing AI systems have developed in parallel with international initiatives. The process (Ichikawa 2020) began in 2016 with the aim of formulating AI principles that could contribute to global rule-making. Since then, Japan has actively promoted AI governance discussions in international forums, including the G7, OECD, and G20. In 2023, under Japan’s presidency, the G7 launched the Hiroshima AI Process, which resulted in the Hiroshima AI Process Comprehensive Policy Framework, one of the first international governance frameworks addressing advanced AI systems. A government official noted (Asahi Shimbun 2024) that Japan leveraged its distinctive position between the innovation-oriented United States and rights-oriented Europe in shaping discussions within the Hiroshima AI Process.

1.2. Democratic Values, Strategic Motivation, and the Role of Private Actors

Across these documents, democratic values are consistently articulated as key governance principles. These include the rule of law, human rights, democracy, diversity, inclusion, fairness, human-centricity, safety, transparency, accountability, security, privacy and personal data protection, fair competition, and digital literacy. Among governmental policies, the Social Principles of Human-Centric AI, adopted in 2019, established the foundational normative framework, emphasizing that AI development and use should respect human dignity and democratic values.

At the same time, it is important to note that AI policy formulation in Japan is driven not solely by the promotion of democratic values but also, and perhaps more strongly, by strategic imperatives to close the gap with leading global AI industries. As highlighted in the AI Act (Cabinet Office, Government of Japan 2025), “Japan is lagging behind in AI development and use,” and the government has accordingly identified AI as a priority area within its national Growth Strategy (Cabinet Secretariat, Government of Japan N.d.a).

Private actors in Japan are also actively engaging with the social implications of AI. Many universities have established ELSI (Ethical, Legal, and Social Issues) centers to study how emerging technologies affect society. Multi-stakeholder initiatives, such as the Hitachi Kyoto University Laboratory and the Kyoto Institute of Philosophy, facilitate interdisciplinary discussion on these challenges. In addition, organizations such as the AI Governance Association and Digital Policy Forum Japan have contributed policy proposals to guide national responses to AI. Together, these activities reflect a broad and multifaceted engagement with the relationship between AI and society in Japan.

2. South Korea

2.1. AI Governance Under G3 Vision

South Korea’s AI governance is founded on the ambitious "AI G3" vision (Ministry of Science and ICT 2025a), a strategic imperative to catch up with the U.S. and China through an innovation-led framework. A central pillar of this approach is the government’s responsiveness to industrial demands (Seo 2025) for deregulation. The private sector frequently argues that compared to global leaders, Korea’s restricted environment hinders competitiveness, leading the state to prioritize industry-friendly policies. Instead of rigid mandates, Korea utilizes the "AI Basic Act"(National Assembly of the Republic of Korea 2026) as a broad legal ground supported by soft law designed to foster a flexible environment. By emphasizing regulatory sandboxes, which function as controlled environments where companies can test AI technologies with relaxed regulatory requirements, the state aims to maximize industrial experimentation while selectively applying oversight to "High-Impact AI," effectively positioning technological capability as the cornerstone of national security and economic growth. The law’s regulatory scope also has extended to foreign AI systems affecting users or markets within South Korea. The Ministry of Science and ICT (MSIT) is empowered to investigate violations and may issue corrective orders or impose adrninistrative fines of up to KRW 30 million (approximately USD 22,000).

2.2. Alteration in Democratic Values

The current Lee adrninistration’s “AI Basic Society” is grounded in its broader policy framework of a “Basic Society,” which emphasizes citizens’ egalitarian access to social services based on principles of redistributive justice. While the previous leadership left a number of statements asserting the protection of liberal democracy under the AI-era as a primary objective, its institutionalization and practical application often remained vague. In contrast, the current adrninistration has adopted a more proactive stance, championing the "AI Basic Society" model where AI literacy and access to AI-assisted social services are framed as universal rights. Beyond domestic policy, Korea is actively asserting its influence in shaping international AI norms through major initiatives such as the AI Seoul Summit[1] (AI Seoul Summit n.d.), following the inaugural AI Safety Summit in the United Kingdom, and a leadership role in the Gyeongju APEC (Korea Policy Briefing 2025).

2.3. Efforts of Civil Society and Academia

While the state drives the technological powerhouse narrative, a robust network of academia and civil society acts as both an enabler and a watchdog for democratic AI development. Groups like Jinbonet and People’s Solidarity for Participatory Democracy monitor the side-effects of emerging AI threats and speak on behalf of those challenged by the asymmetric strength of big-tech incorporations. Furthermore, social organizations like Parti have endeavored to redefine deliberative democracy by using AI to function as an instrument for citizens to deliver mass opinions into coherent policy agendas. Simultaneously, university-based centers such as Seoul National University AI Institute’s Center for Industry Collaboration have provided the theoretical rigor for trustworthy AI, ensuring democratic values are not sidelined in policy drafts. Despite these efforts, a recurring challenge remains. Civil society participation and academic discussion often lack tangible veto power over industrial strategies, leaving the final direction of AI development largely in the hands of the state and major tech firms.

3. Taiwan

Taiwan is characterized by a combination of high vulnerability and an emerging form of social resilience: it is a democracy exposed to intense information pressure from the PRC, yet one in which efforts to resist such pressure have also taken shape. Despite ongoing concerns over Chinese electoral interference, disinformation, and influence operations, Freedom House 2025 (Freedom House 2025) continues to rate Taiwan highly on both political rights (38/40) and civil liberties (56/60). In this sense, Taiwan can be understood as a society that remains highly exposed to PRC-originated foreign information manipulation and interference (FIMI) while continuing to sustain democratic institutions. This resilience is supported by a dense infrastructure of civil society and digital participation. The Taiwan Open Government National Action Plan（2021–2024) &（2025–2028) institutionalizes co-governance between the state and civil society and sets out concrete commitments to transparency, accountability, and expanded citizen participation (National Development Council 2021a; 2025). Alongside this framework, Taiwan’s civic and digital-democratic ecosystem includes g0v[2], JOIN[3], and vTaiwan[4], while the integrity of the information environment is supported by organizations such as the Taiwan FactCheck Center (TFC)[5], MyGoPen[6], FactLink[7], Doublethink Lab[8], and Taiwan Information Environment Research Center (IORG)[9], which maintain a degree of independence from the government even as they interact with public institutions.

At the institutional level, Taiwan’s Artificial Intelligence (AI) Basic Act (Legislative Yuan of the Republic of China 2026) was passed by the Legislative Yuan in December 2025 and came into effect in January 2026. The Act seeks to harmonize technological progress with social welfare and sets out basic principles including human autonomy, privacy and data governance, safety, transparency, fairness and non-discrimination, and accountability. Although it does not itself impose penalties, Taiwan appears to be adopting a regulatory approach that combines a general principles-based law with more specific sectoral policies to address particular risks. Under the current law, the National Science and Technology Council[10] is designated as the central competent authority and is responsible for AI research and talent development, while the Ministry of Digital Affairs (MODA)[11], established in 2022 to integrate digital policy across telecommunications, information, communication, cyberspace, and the internet, has become the de facto driver of the Basic Act’s implementation.

At the same time, Taiwan positions AI not only as an object of regulation but also as a core component of national growth strategy. According to the National Development Council (NDC)[12], the Ten AI Initiatives Promotion Plan (2025-28) was approved by the Executive Yuan in January 2026 and presented as a national strategy to transform Taiwan into a Smart Technology Island (National Development Council 2026). The plan seeks to generate NT$15 trillion (about US$474 billion) in economic value and create 500,000 high-paying jobs by 2040, with sovereign AI, digital and computing infrastructure, advanced technologies, and the establishment of three international-level laboratories forming its central pillars (Taiwan Today 2026). Taiwan can therefore be seen both as a democracy that is highly vulnerable to PRC-originated FIMI and as an advanced case of democratic self-defense in the AI era, supported by civil society and institutional innovation.■

[1] AI Seoul Summit. "AI Seoul Summit 2024." https://aiseoulsummit.kr/aiss/.

[2] g0v. "g0v: Decentralised Civic Tech Community, Taiwan." https://g0v.tw/intl/en/.

[3] National Development Council (Taiwan). "Join: Public Policy Network Participation Platform." https://join.gov.tw/.

[4] vTaiwan. "vTaiwan: Rethinking Democracy." https://info.vtaiwan.tw/.

[5] Taiwan FactCheck Center. "Home." https://en.tfc-taiwan.org.tw/.

[6] MyGoPen. "MyGoPen: Fact-Checking and Rumor Clarification Platform, Taiwan." https://www.mygopen.com/.

[7] FactLink. "FactLink 數位素養實驗室: Digital Investigation, AI Literacy, Media Literacy." https://www.factlink.tw/.

[8] Doublethink Lab. "Doublethink Lab: Strengthen Democracy Through Enhancing Digital Defenses." https://doublethinklab.org/.

[9] IORG (Taiwan Information Environment Research Center). "About IORG." https://iorg.tw/_ua/about.

[10] National Science and Technology Council, Republic of China (Taiwan). "National Science and Technology Council, R.O.C." https://www.nstc.gov.tw/?l=en.

[11] Ministry of Digital Affairs (Taiwan). "Ministry of Digital Affairs." https://moda.gov.tw/en/.

[12] National Development Council (Taiwan). "National Development Council." https://www.ndc.gov.tw/.

Digital Democracy

Digital democracy initiatives have developed along distinct trajectories, reflecting different understandings of how digital technologies should support democratic governance. In Japan, efforts have primarily focused on using digital tools, including AI, to visualize citizens’ opinions and facilitate consensus-building in policy discussions. Initiatives have emerged across the private sector, local governments, and political actors, but they remain largely experimental and project-based, with limited impact on formal national decision-making processes. In contrast, South Korea has pursued a more systemic and institutionalized trajectory, focusing on upgrading its established participatory platforms with AI-driven facilitators to enhance policy co-design and creating high-level bodies like the AI Democracy Committee. Taiwan represents another trajectory, often highlighted as a model of digital democracy built upon a strong civic-tech ecosystem with institutionalized channels for public participation like vTaiwan. At the same time, each approach faces limitations: Japan’s initiatives remain experimental, Korea’s regulatory approach raises concerns about potential restrictions on political expression, and Taiwan’s participatory platforms have been criticized for engaging only a relatively narrow segment of society.

1. Japan

1. Increasing Utilization by Private Actors and Local Governments

Initiatives related to digital democracy in Japan can be observed across the private sector, local governments, and the political sphere. Practical experiments and pilot efforts have increased, but arrangements in national decision-making remain limited.

In the private sector, initiatives have emerged that collect citizens’ views and provide opportunities for discussion, and some tools are already used in adrninistrative and project settings. Digital Democracy 2030[1] presents a vision of using digital technologies to make citizens’ views visible and to inform political and adrninistrative processes. Liquitous[2] has been used to support online discussion and consensus-building in specific projects, including collaboration with local governments.

Local governments, particularly the Tokyo Metropolitan Government, have also advanced related efforts. GovTech Tokyo[3] has promoted adrninistrative digitalization and supported data use and operational capacity across the metropolitan government and municipalities. In addition, Tokyo has applied broad listening methods (Tokyo Metropolitan Government, Office of the Governor for Policy Planning 2025) in the formulation of its long-term policy agenda. In the development of the Tokyo 2050 strategy, tools associated with the Digital Democracy 2030 initiative were used to collect and analyze a wide range of citizens’ views. GovTech Tokyo supported this process by building systems to gather digital inputs and by using AI to visualize and organize the collected opinions. These practices were intended to inform policy discussions rather than directly determine outcomes.

1.2. Limited Realization in National Politics

In the political sphere, digital approaches have begun to be discussed but have not yet become institutionalized. The Tokyo gubernatorial election provided an early example in which digital tools were used to engage with citizens and communicate policy positions. These efforts were associated with initiatives led by Anno Takahiro, AI engineer and Member of the House of Councillors of Japan, and related actors who sought to incorporate digital methods into political engagement. Led by Anno, Team Mirai increased its number of seats in the Diet and has sought to use AI to understand public views and support policy development. Cross-party study groups on AI have been established to share knowledge and discuss policy issues, but these efforts remain at the stage of learning and agenda formation rather than institutional practice.

A range of actors are involved and practical initiatives are increasing. At the same time, these efforts are often organized as individual projects, and their connection to formal decision-making remains limited, particularly at the national level. The AI Basic Plan states that the government will take a leading role in “actively and proactively utilizing AI” to promote broader usage across the general population. In line with this objective, the Digital Agency has launched (Matsumoto 2026) a large-scale demonstration project for the government AI system. While these developments indicate that the use of AI within bureaucratic organizations has begun to expand, they have not yet reached a stage where they fundamentally reshape democratic decision-making processes.

2. South Korea

2.1. Evolution of Civic Engagement and Democracy

South Korea is evolving its developed digital legacy by enhancing established citizen participatory platforms, from e-People[4] to Sotong24[5], into a more AI-driven participatory governance model (Anti-Corruption and Civil Rights Commission 2026). Rather than replacing these successful foundations, the government is integrating AI as an intelligent facilitator to more efficiently classify, summarize, and respond to the massive inputs of citizen voices, thereby reinforcing the role of citizens as active policy co-designers. Simultaneously, to address concerns over the black-box nature of AI and the potential exclusion of specific social groups, the state is institutionalizing ethical safeguards, exemplified by the “Ethical Guidelines for AI in the Public Sector” (Ministry of the Interior and Safety 2025) at the national level and the “Seoul AI Ethics Guidelines” (Seoul Metropolitan Government 2026) at the regional level. Beyond fostering participation, the government elevates democratic principles to a core priority within the national AI strategy. As a pivotal move, the AI Democracy Subcommittee (Ministry of Science and ICT 2026b) was recently added to the existing subcommittees of the Presidential Council on National Artificial Intelligence Strategy, marking a monumental decision to institutionalize democratic values at the highest level of AI leadership.

2.2. Democratic Equality and Inclusive Legislation

South Korea is intensifying legislative efforts to ensure technological benefits are shared within all nations equally. The "Digital Inclusion Act" (National Assembly of the Republic of Korea 2025) and proposed amendments to the "AI Framework Act" (Ministry of Science and ICT 2025d) specifically aim to protect vulnerable populations, codifying their rights to AI accessibility and participating in AI governance. These legal frameworks focus on preventing social exclusion by mandating protections for the marginalized. Complementing these policies, Big Tech is also deploying AI for universal social welfare. For instance, Naver’s "CLOVA CareCall"[6] utilizes generative AI to monitor the health of the elderly, demonstrating how technology can serve as a practical tool for protecting vulnerable citizens.

Korea demonstrates a willingness to embed democratic safeguards directly into its AI governance blueprint, characterized by institutional proactivity ranging from an AI-driven participatory governance model to the establishment of the AI Democracy Subcommittee inside the AI Presidential Council. While the Digital Inclusion Act and corporate initiatives like CLOVA CareCall reflect a genuine legislative and corporate push toward inclusive AI, the gap between codified rights and actual accessibility for vulnerable populations remains a critical variable yet to be fulfilled. Ultimately, while Korea’s democratic AI framework is theoretically sound, its long-term credibility will depend on whether these institutional designs translate into meaningful civic participation, with no groups excluded.

3. Taiwan

3.1. The Concept of Taiwan’s Digital Democracy: “Plurality”

Taiwan’s model of digital democracy is grounded in the idea of plurality[7], advocated by figures such as Audrey Tang (Tang 2025), who became MODA’s inaugural minister upon its establishment in 2022. Rather than relying on simple majoritarianism, the plurality approach seeks to avoid the polarization that majority rule can generate. Even if 51 percent support a decision, the dissatisfaction of the remaining 49 percent may endure, fostering backlash and distrust. Taiwan’s approach, by contrast, aims not to erase differences but to identify overlap across diverse views and use it to advance democracy. Through open-source tools such as Pol.is[8] and Talk to the City (T3C)[9], participatory platforms such as vTaiwan[10] and JOIN[11], and close coordination with government institutions—particularly the NDC[12]—Taiwan has sought to combine broad public participation with deeper deliberation.

3.2. Open Government and Civil Society

Taiwan’s open-government agenda, including open data initiatives, entered a more formal and institutionalized phase with the adoption of a national action plan in 2021. Overseen by the NDC, the Taiwan Open Government National Action Plan（2021–2024) &（2025–2028) institutionalizes co-governance between the state and civil society and sets out concrete commitments to transparency, accountability, and expanded citizen participation (National Development Council 2021a; 2025). In practice, these principles have been operationalized through a broader ecosystem of civic-tech initiatives and participatory infrastructures, including g0v, vTaiwan, JOIN, and, more recently, T3C.

g0v[13] is a decentralized civic-tech community launched in 2012. Based on transparency and open collaboration, g0v has addressed public issues from outside the government. A representative example is Cofacts[14], a collaborative fact-checking system that enables citizens to verify suspicious information circulating on LINE, the messaging app dominant in Taiwan.

vTaiwan[15] is an online-offline public consultation process launched in 2014 to enable citizens, government officials, businesses, and experts to deliberate on national policy issues. Using open-source tools such as Pol.is[16], participants can register agreement, disagreement, or uncertainty, making it possible to visualize large-scale opinion distributions and identify areas of potential consensus. Rather than amplifying conflict, vTaiwan is designed to surface bridge-building arguments and actionable common ground. By 2023, it had handled more than 28 cases, around 80 percent of which reportedly led to concrete government action. Representative cases include deliberations over Uber Regulation (Tang 2016) in 2015 and the FinTech framework introduced in 2018.

JOIN[17] is the NDC’s official online participation platform, established in 2015[18] as a permanent channel through which citizens can submit opinions, make policy proposals, and monitor implementation. Under the Directions for Implementing Online Participation in Public Policy (Executive Yuan, Republic of China (Taiwan) 2018), the NDC serves as the platform’s managing authority. Proposers are defined as persons with Taiwanese nationality or permanent residency in Taiwan, and no explicit age requirement is stipulated (Point 4(1)). If a proposal gathers 5,000 supporting signatures within 60 days, the competent authority must publish a concrete response within two months (Point 7 (2)). According to the NDC (National Development Council 2021b), from the platform’s launch in 2015 to 31 January 2021, JOIN received 10,411 proposals in total; 5,248 advanced to the co-signature stage, 233 became formal cases, 220 received official responses, and 105 were adopted by government agencies. In 2024 alone, the National Council for Sustainable Development (National Council for Sustainable Development 2025) reported 2,801 proposals, of which 1,236 advanced to the co-signature stage, 24 became formal cases, and 10 were partially adopted. One illustrative case is the 2018 proposal calling for legal access to artificial insemination and IVF for single women (Wu 2018). After gathering 5,124 supporting signatures within 60 days, it received a formal response from the Ministry of Health and Welfare and was connected to a collaborative consultation process. In 2025, a draft amendment to the Assisted Reproduction Act (Ministry of Health and Welfare 2025), extending access to unmarried women and women in same-sex marriages, passed the Executive Yuan.

T3C[19], released as open source in October 2023, is an AI analysis tool that uses LLMs to process large volumes of open-ended responses and deliberative data. According to its report[20], its key innovation lies in summarizing large-scale public input while preserving the nuance and diversity of views. In collaboration with MODA, T3C has been used to analyze large response datasets on issues such as the AI Assembly 2023 workshops[21] and post-election political analysis of the Democratic Progressive Party (DPP) in Taiwan’s 2024 presidential election.[22]

3.3. Critical Evaluation

Overall, Taiwan’s civil society appears comparatively strong, and Glen Weyl[23] identifies Taiwan as a leading example of the intersection of technology and democracy. Audrey Tang’s international prominence has further elevated the visibility of this model. At the same time, however, JOIN remains limited in reach and tends to attract a relatively small and elite segment of society, while the continued citation of vTaiwan as a flagship case, despite being more than a decade old, suggests that Taiwan’s digital democracy is still often understood through one of its earliest and most visible cases. As Chihhao Yu[24] of g0v argues, digital tools should be seen as complementary to, rather than constitutive of, democracy, whose foundation ultimately lies in offline deliberation and civil society.■

[1] Digital Democracy 2030 Project (デジタル民主主義2030). "Home." https://dd2030.org/.

[2] Liquitous. "Liquitous: Civic Participation and Consensus-Building Platform." https://liquitous.com/.

[3] GovTech Tokyo (GovTech東京). "GovTech Tokyo." https://www.govtechtokyo.or.jp/.

[4] Anti-Corruption and Civil Rights Commission (Korea). "e-People: National Civil Petition Service." https://www.epeople.go.kr/index.jsp.

[5] Ministry of the Interior (Korea). "Sotong24: Online Public Communication Platform (소통24)." https://sotong.go.kr/front/main/index.do.

[6] NAVER Cloud. "CLOVA CareCall: AI Welfare Call Service." https://www.ncloud.com/product/aiService/clovaCareCall.

[7] Weyl, E. Glen, Audrey Tang, and the Plurality Community. "Plurality: The Future of Collaborative Technology and Democracy." https://plurality.net/.

[8] The Computational Democracy Project. "Polis." https://pol.is/signin.

[9] AI Objectives Institute. "Talk to the City." https://talktothe.city/.

[10] vTaiwan. "vTaiwan: Rethinking Democracy." https://info.vtaiwan.tw/.

[11] National Development Council (Taiwan). "Join: Public Policy Network Participation Platform." https://join.gov.tw/.

[12] National Development Council (Taiwan). "National Development Council." https://www.ndc.gov.tw/.

[13] g0v. "g0v: Decentralised Civic Tech Community, Taiwan." https://g0v.tw/intl/en/.

[14] Cofacts. "Cofacts: Message Reporting Chatbot and Crowd-Sourced Fact-Checking Community." https://en.cofacts.tw/.

[15] vTaiwan. "vTaiwan: Rethinking Democracy." https://info.vtaiwan.tw/.

[16] The Computational Democracy Project. "Polis." https://pol.is/signin.

[17] National Development Council (Taiwan). "Join: Public Policy Network Participation Platform." https://join.gov.tw/.

[18] National Development Council (Taiwan). "About Join: Public Policy Network Participation Platform." https://join.gov.tw/aboutus/index/en_US.

[19] AI Objectives Institute. "Talk to the City." https://talktothe.city/.

[20] AI Objectives Institute, "Talk to the City," https://talktothe.city/.

[21] AI Objectives Institute, "AI Assembly 2023 Workshops," Talk to the City, 2023, https://talktothecity.org/report/ai-assembly-2023-workshops_1-translations.

[22] AI Objectives Institute, "Talk to the City," https://talktothe.city/. For a case study of T3C deployment in Taiwan, see AI Objectives Institute, "Taiwan 2024 DPP Party – Post Presidential Election Politics Analysis (In Chinese)," https://talktothecity.org/report/taiwan-2024-dpp-ZH.

[23] Glen Weyl, interview by TBS CROSS DIG with Bloomberg, "Japan and Taiwan's Democracy and Digital Innovation (In Japanese)," YouTube, May 31, 2025, https://www.youtube.com/watch?v=gTWqbdfTc2g.

[24] Yu, Chihhao. "Chihhao Yu: Profile." Global Taiwan Institute. https://globaltaiwan.org/member/chihhao-yu/.

Information Integrity

Japan, South Korea, and Taiwan all face growing pressure on the integrity of the information space, especially from AI-enabled disinformation and foreign influence operations, but they differ in the nature of the threat, the form of response, and the main sources of vulnerability. In Japan, AI-generated disinformation and biased content are increasingly seen as risks to social stability and national security, yet legal responses remain systern and continue to rely largely on voluntary measures and self-regulation because of the need to balance countermeasures with freedom of expression. In South Korea, institutional responses have been more rapid and coordinated, but the weakening of independent fact-checking has emerged as a major structural vulnerability. Although there are some signs of possible reconstruction, the broader fact-checking ecosystem has not yet recovered. In Taiwan, PRC-linked foreign information manipulation constitutes the central threat, and responses have combined legal reforms with extensive civil society efforts in fact-checking, prebunking, research, and media literacy. Even so, these initiatives remain constrained by insufficient platform regulation and inadequate funding for fact-checking.

1. Japan

1.1. Risk Recognition

AI-generated disinformation, misinformation, and biased content are recognized as risks that may affect decision-making and destabilize society. In Japan, these risks have been acknowledged in government guidelines, including the AI Business Guidelines, which highlight[1] concerns related to social stability, human rights, and cultural diversity.

1.2. Countermeasures and Their Tension with Freedom of Expression

Countermeasures have been developed mainly through voluntary efforts. In its 2020 concluding report, a study group on digital platforms established in 2016 under the Ministry of Internal Affairs and Communications emphasized (Ministry of Internal Affairs and Communications 2020) the role of voluntary measures by private actors in addressing online content issues, reflecting a systern approach to government intervention in light of freedom of expression. This direction contributed to the establishment of the Japan Fact-Check Center (JFC). Legislation also reflected the government’s reluctance to intervene. The Information Distribution Platform Act, enacted in 2024, requires platform operators to respond more promptly to rights-infringing content and to ensure transparency in content removal standards and their implementation, but its scope is confined to illegal or harmful content and does not fully address disinformation.

Policy discussions to advance countermeasures have faced difficulties. A study group established in 2023 under Ministry of Internal Affairs and Communications (MIC) proposed (Ministry of Internal Affairs and Communications 2024) measures such as content moderation and advertising regulation, but Nikkei reported (Sakai 2024) that the group was dissolved as coordination became increasingly complex, particularly in balancing countermeasures with freedom of expression. A subsequent group suggested (Ministry of Internal Affairs and Communications 2025b) clarifying illegal information and encouraging voluntary codes of conduct by platforms, but concrete regulatory measures remain underdeveloped. Asahi reported (Wakae 2025) that the final report ultimately prioritized non-binding “voluntary codes of conduct,” and that even a previously suggested pathway toward stricter regulation was removed. Reportedly, this shift reflects both the difficulty of drawing boundaries in relation to freedom of expression and broader political constraints, including concerns about U.S. opposition to platform regulation.

In contrast, traditional media organizations have begun to take a more proactive role. Amid the increasing spread of disinformation and misinformation, they have become more active in fact-checking during elections since 2025. Historically, many outlets avoided fact-checking to preserve neutrality. Consequently, 2025 is often described (NHK 2025) as the first year in which systematic fact-checking was widely implemented during elections, while the impact is likely to be limited and may even inadvertently contribute to increased public exposure to misinformation.

1.3. National Security and Technological Approaches

At the same time, responses from a national security perspective have developed in parallel. Disinformation was recognized (Cabinet Secretariat of Japan 2022) as a national security issue in the National Security Strategy in 2022, and efforts have been made across ministries. The government has also moved to (Kyodo News 2026) strengthen intelligence functions, and it has reported that a specialized unit addressing disinformation has been established. However, existing measures remain limited in scope. For example, the Ministry of Defense has primarily focused (Ministry of Defense of Japan N.d.) on publishing fact-checking information related to disinformation targeting itself

Technical responses have also been promoted. MIC has supported (Ministry of Internal Affairs and Communications N.d.) the development of technologies to detect and mitigate disinformation. The AI Safety Institute (AISI) has developed evaluation frameworks (Japan AI Safety Institute 2025a) that incorporate risks such as harmful content, misinformation, bias, fairness, and inclusiveness, and has also issued guidance on red-teaming (Japan AI Safety Institute 2025b) from an adversarial perspective. Private-sector collaborations, such as Frontria, have further contributed to these efforts.

Despite these initiatives, current responses remain insufficient relative to the scale of AI-driven information risks. Existing measures are fragmented across voluntary initiatives, limited regulatory frameworks, and security-oriented responses, and have yet to form a comprehensive approach to safeguarding the integrity of the information environment.

2. South Korea

2.1. Crisis of Independent Fact-Checking

South Korea’s independent fact-checking ecosystem is facing a critical contraction. The SNU FactCheck Center, previously the nation's central verification hub, entered an indefinite pause (Kim 2024) in 2024 following the withdrawal of financial support from Naver (Lim 2023). This followed the 2023 dissolution of FactCheckNet (Ju 2025), which suffered from both public budget cuts and intense political accusations of partisan bias from conservative circles. Consequently, neutral, third-party verification has weakened, as increasing political polarization makes maintaining a non-partisan truth-seeking ecosystem financially and socially difficult. While these setbacks have weakened neutral verification, a new momentum for revival is surfacing. The Korea Media and Communications Commission (KMCC) is currently spearheading the establishment of a “Transparency Center” aimed at reactivating private-sector fact-checking and ensuring the reliability of digital information (Korea Communications Commission 2025).

2.2. Regulatory Legislation and Multi-Agency Response

South Korea has implemented a series of measures to counter AI-generated disinformation, responding to the rapid evolution of the technology. The government introduced an AI-generated content labeling policy (Lee 2025) as an initial step. This was reinforced by the KMCC’s launch of a 24-hour fast-track system in late 2025 (Lee 2025), which prioritized the swift review and adrninistrative blocking of deceptive AI advertisements. Countering disinformation is further expanded to protecting democratic values, especially to maintaining neutral settings for ballots. Following the 2023 enactment of the Public Official Election Act, which stipulated a 90-day pre-election ban on deepfakes, the National Election Commission (NEC) adopted a specialized AI model designed to detect deepfake content. “Aegis” was developed by National Forensic Service and Korea Electronics Technology Institute in order to safeguard the electoral environment. In February 2026, the scope of response was broadened with a pan-governmental meeting involving MSIT, the Ministry of the Interior and Safety, the Ministry of Justice, and KMCC to strengthen enforcement and penalties (Korea Communications Commission 2026). In parallel, public-private cooperation has intensified as well, such as AI Safety Institute and Kakao co-developing real-time detection tools (Jang 2025). These tools are slated for open-source distribution to the NEC and the private sector to enhance collective resilience against disinformation.

While Korea has generated significant regulatory momentum through the rollout of AI content labeling, KMCC’s fast-track system, and pan-governmental coordination, the long-term effectiveness of these measures against evolving disinformation remains to be proven. This institutional progress is currently tested by a deeply embedded structural vulnerability, such as the collapse of the nation's fact-checking infrastructure, driven by sustained political attacks on its legitimacy and a total withdrawal of financial support. Although the KMCC’s Transparency Center offers a potential path for revival, the precarious conditions that caused the original collapse persist, leaving the long-term sustainability of the ecosystem uncertain. Ultimately, while Korea demonstrates a strong capacity for top-down adrninistrative coordination, sustaining a truly pluralistic and independent verification ecosystem remains a major hurdle that depends as much on a stable political and financial environment as it does on technological solutions.

3. Taiwan

3.1. Threat Structure and Recent Trends

The principal threat to the integrity of Taiwan’s information space can be identified as influence operations originating from the PRC. According to Tim Niven[2], the PRC’s FIMI targeting Taiwan operates through a three-layered structure. At the first level are strategic objectives, including the promotion of unification, the erosion of trust in democracy, and the creation of anxiety about Taiwan’s future. At the second level are deeper narratives, such as claims that the government is corrupt, elections are rigged, the military is weak, and the United States would not come to Taiwan’s aid. At the third level are granular stories—such as failures in military training or scandals involving political candidates—which are presented as if they were concrete evidence substantiating those broader narratives.

A further important point is that the character of the PRC’s information operations has changed in recent years; earlier propaganda more directly emphasized the supposed appeal of unification (Niven 2023). However, since support for unification remains low in Taiwan, the focus has increasingly shifted toward polarizing Taiwanese society and weakening democracy from within.

3.2. The Intensification of Threats through Social Media and AI

One important channel through which FIMI operates in Taiwan is the amplification of narratives through influencers and other actors embedded within Taiwanese society, rather than solely through propaganda imposed from outside. As Tim Niven of Doublethink Lab explained in an interview, the PRC may support the circulation of pro-PRC narratives through local collaborators and influencers, both before and after dissemination, including through financial support. On platforms such as TikTok, which are especially influential among younger audiences, a “propaganda economy” can emerge in which pro-PRC narratives are amplified through attention, engagement, and monetization. Although no direct financial link to the PRC has been publicly confirmed, the case of the China-born influencer yaya in Taiwan illustrates the issue; after publicly advocating military unification, her dependent residency permit was revoked and she was ordered to leave Taiwan (National Immigration Agency 2025).

A second major concern is the way AI degrades the information environment more broadly, beyond the spread of individual falsehoods. As Summer Chen[3] of FactLink noted in an interview, AI is used in at least four ways. First, it enables direct fabrication, such as fake videos or audio recordings of politicians. Second, it produces AI-generated mini clips (Li, Chen, and Ma 2025), propaganda that provokes anger, anxiety, or ridicule; even in the absence of clear factual claims, such content can distort political perception. Third, it generates large volumes of low-quality “AI slop,” often circulated through short-form video platforms, which undermines concentration, crowds out credible information, and degrades the quality of public deliberation. Fourth, it facilitates the operation of fake accounts and other forms of automated information manipulation. As Summer noted, people may recognize that such content is AI-generated and untrue, yet still consume it with little concern. As a result, they become more susceptible to anger and other affective reactions, making reasoned public debate more difficult and rendering democracy more vulnerable.

3.3. Government Responses

In response to these challenges, the Taiwanese government has increasingly relied on legal and institutional measures. In particular, between late 2022 and 2023, amendments to the Public Officials Election and Recall Act (Public Officials Election and Recall Act 2026) and the Presidential and Vice-Presidential Election and Recall Act (Presidential and Vice-Presidential Election and Recall Act 2023) strengthened the legal framework for responding to election interference involving deepfakes and expanded regulation of election-related content on digital platforms. These developments build on the Anti-Infiltration Act (Anti-Infiltration Act 2020), which came into force in 2020 and prohibits political and election-related activities conducted under the direction, commission, or financial support of foreign hostile forces. In addition, Article 53 of the Disaster Prevention and Protection Act (Disaster Prevention and Protection Act, Art. 53) addresses false information during disasters, while Article 63(1)(5) of the Social Order Maintenance Act (Social Order Maintenance Act 2025) addresses rumors that disturb public order. MODA has also warned the public against downloading unnecessary Chinese apps and platforms such as TikTok, Douyin, and RedNote (Chiu and Chung 2025).

3.4. Civil Society Responses

Alongside these government measures, civil society has developed a diverse range of responses. Some organizations focus primarily on debunking and fact-checking. TFC[4], which was co-founded in 2018 by the Taiwan Media Watch Foundation and the Association for Quality Journalism, relies primarily on claim-by-claim debunking—that is, the verification and refutation of individual claims. At the same time, it engages in international cooperation through initiatives such as the Youth Verification Challenge[5]. MyGoPen[6] provides verification services directly to the public through LINE-based responses and chatbot functions. Both TFC and MyGoPen have benefited from support through Meta’s Third-Party Fact-Checking Program (Li and Chen 2025).

Other civil society actors place greater emphasis on prebunking, narrative analysis, and research on the broader structure of information manipulation. As Summer Chen of FactLink[7] explained in an interview, FactLink[8] pays greater attention to the broader narratives that underpin recurring patterns of PRC disinformation, rather than focusing on individual cases. To this end, it develops narrative-based playbooks (Li, Chen, and Ma 2026), collaborates with specialists on Open-Source Intelligence (OSINT) tools, publishes AI image verification guidelines (FactLink 2025), and provides capacity-building for journalists and educators. Doublethink Lab[9], a Taiwanese civil society research organization focused on the PRC’s digital influence operations, also engages in international collaboration. According to Tim Niven in an interview, it uses agentic AI to partially automate the identification of fake accounts and media networks. IORG[10] likewise plays an important role in researching and analyzing PRC information operations and Taiwan’s broader information environment.

3.5. Future Prospects

Taken together, Taiwan’s responses to information threats can be characterized as a distributed model supported by government institutions, civil society organizations, researchers, and international networks. While the government has strengthened legal frameworks in recent years, institutional responses to influencer-driven propaganda economies and affective content remain insufficient.

At the same time, civil society actors face severe financial constraints, as well as a broader decline in the legitimacy and public acceptance of fact-checking itself. As Huian Ho of TFC[1] noted in an interview, Meta’s January 2025 announcement (Kaplan 2025) that it would end cooperation with third-party fact-checkers in the United States generated anxiety within Taiwan’s fact-checking community and prompted budget cuts as organizations prepared for future uncertainty. Despite these challenges, Taiwanese organizations continue to build information integrity coalitions through collaboration with domestic experts, journalists, educators, and international partners. These efforts constitute an important foundation for strengthening resilience.■

[1]Ministry of Economy, Trade and Industry and Ministry of Internal Affairs and Communications, Japan. 2025. "AI Guidelines for Business (Ver. 1.1): Appendix." April 4. https://www.soumu.go.jp/main_content/001003029.pdf (Accessed: April 5, 2026).

[2] Niven, Tim. "Tim Niven, PhD: Google Scholar Profile." Google Scholar. https://scholar.google.com/citations?user=B9vqlpwAAAAJ&hl=en.

[3] Chen, Summer. "Summer Chen: Profile." Global Taiwan Institute. https://globaltaiwan.org/member/summer-chen/.

[4] Taiwan FactCheck Center. "Home." https://en.tfc-taiwan.org.tw/.

[5] Niven, Tim. "Tim Niven, PhD: Google Scholar Profile." Google Scholar. https://scholar.google.com/citations?user=B9vqlpwAAAAJ&hl=en.

[6] MyGoPen. "MyGoPen: Fact-Checking and Rumor Clarification Platform, Taiwan." https://www.mygopen.com/.

[7] Chen, Summer. "Summer Chen: Profile." Global Taiwan Institute. https://globaltaiwan.org/member/summer-chen/.

[8] FactLink. "FactLink 數位素養實驗室: Digital Investigation, AI Literacy, Media Literacy." https://www.factlink.tw/.

[9] Doublethink Lab. "Doublethink Lab: Strengthen Democracy Through Enhancing Digital Defenses." https://doublethinklab.org/.

[10] IORG (Taiwan Information Environment Research Center). "A Guide to Information Literacy: Exploration and Survival in Taiwan's Information Environment." https://iorg.tw/book.

[11] Ho, Huian. "Huian Ho: Speaker Profile." NECE – Networking European Civic Education. https://nece.eu/speaker/huian-ho/.

Cyber Security

Japan, South Korea, and Taiwan all place growing emphasis on cybersecurity, but they differ in their principal threat perceptions, responses, and remaining weaknesses. In Japan, the threat picture is broad, including cyberattacks on critical infrastructure and AI-related risks, and the response has become more government-led through the Active Cyber Defense framework and the 2025 Cybersecurity Strategy, albeit with concerns over expanded state powers and the secrecy of communications. In South Korea, the threat is centered primarily on North Korea, and the government has responded through its 2024 National Cybersecurity Strategy, which emphasizes offensive defense and stronger private-sector participation, but a weak legal foundation and fragmentation across ministries remain major constraints. In Taiwan, the principal threat comes from the PRC, where cyberattacks are understood as part of a broader hybrid threat environment combining infrastructure disruption with disinformation and other gray-zone tactics; accordingly, Taiwan has strengthened its cybersecurity legislation and resilience-oriented governance, though it continues to face unusually intense and persistent cross-domain pressure.

1. Japan

1.1. Expanding Role of the State in Cybersecurity

Japan is strengthening state capacity in cybersecurity while defining procedural limits on information use as a democratic society. In this context, AI is framed as something to be protected, a tool to be leveraged, and an emerging source of threat.

The enactment of legislation on active cyber defense in May 2025 marked an important institutional development (Cabinet Secretariat of Japan N.d.b). This permits early detection and neutralization of serious cyber threats under specified conditions. Police and the Self-Defense Forces are responsible for implementing neutralization. The policy expands the operational role of the state in cyberspace. At the same time, the framework raised questions concerning the constitutional protection of the secrecy of communications, which had allegedly hindered effective countermeasures (Tsuchiya 2025). Government advisory proposals (Cabinet Secretariat of Japan 2024) and parliamentary materials (Kakinuma 2025) highlighted the need to reconcile cybersecurity measures with constitutional protection of the secrecy of communications. The discussion focused on how communication data could be obtained and used under defined legal requirements, as well as on the design of oversight mechanisms and accountability procedures.

1.2. AI in Cybersecurity

Japan revised its Cybersecurity Strategy in December 2025 and clarified the role of the state in responding to cyber threats (Cybersecurity Strategic Headquarters 2025). The revised strategy assigns the national government a stronger coordinating function and introduces active cyber defense as a central element of policy. The strategy situates AI within cybersecurity policy in three ways: securing AI systerns (Security for AI), using AI to improve defensive capacity (AI for Security), and addressing attacks that misuse AI. Policy measures include research and development, guideline formation, and international coordination. The government has stated that it will work with the AI Safety Institute and promote international processes such as the Hiroshima AI Process. It also requires government agencies to manage risks when procuring and using generative AI.

AI has also been discussed in the context of economic security policy. The Economic Security Promotion Act enacted in May 2022 includes support for the development of critical technologies, including cloud programs, as one of its pillars. Based on the Act, to secure domestic computational resources for generative AI, the Ministry of Economy, Trade, and Industry has provided subsidies for private-sector investment in domestic cloud and computing infrastructure (Ministry of Economy, Trade and Industry of Japan N.d.). Discussions on revisions to the economic security framework have also addressed the protection of data centers and cloud services including measures to prevent external interference, information leakage, and data loss, as well as mechanisms for the government to monitor the location and status of data center facilities.

2. South Korea

2.1. Legislative and adrninistrative Stagnation

South Korea's cybersecurity posture is currently defined as bold strategic ambition and a persistent legislative stagnation. While the 2024 National Cybersecurity Strategy (Office of National Security 2024) advocates for a bold "offensive defense," the nation’s international standing remains surprisingly low on the Cybersecurity Strategy Scorecard (Heiding, O'Neill, and Price 2025) compared to global peers. This lag is primarily attributed to a decade-long delay in policy formulation and a lack of sustained public engagement and communication. Intensifying this situation, the National Cybersecurity Basic Act has remained in a state of legislative paralysis since 2006 (Kim 2024). This deadlock stems from intense inter-agency power struggles for organizational leadership and dominance, leaving a critical void in the unified legal framework required for effective adrninistrative execution (Shin 2025).

2.2. Countering North Korean Threats through Private-Public Synergy and AI

Faced with an existential threat environment where North Korean actors accounted for approximately 80% of public cyberattacks (KBS 2024), Korea is rapidly evolving its defense paradigm. These attacks have moved beyond individual targets to strike core national infrastructure, including judicial and adrninistrative networks. In response, private sector engagement is becoming increasingly active, moving away from a state-mandated approach. Collaborative frameworks such as the Korea Cyber Security Alliance (Bang 2025) and Information Security Companies Autonomous Security Council (Jo 2025) now facilitate real-time threat intelligence sharing between the major industries and government. This evolution is further accelerated by AI-driven innovation, illustrated by the 2025 AI Cyber Defense Contest (Korea Internet & Security Agency 2025).

While Korea’s 2024 strategy targets a bold "offensive defense," it currently lacks a concrete operational blueprint, relying on aspirational language without the necessary budgets or specific agency assignments to ensure implementation. This strategic void is further widened by the legislative paralysis surrounding the National Cybersecurity Act, which remains stalled due to intense inter-agency power struggles. Conversely, the persistent threat from North Korea is forcing practical progress where legislation remains lacking, fostering active private sector engagement out of necessity. However, without a unified legal framework, these efforts remain fragmented across various ministries, leaving the systern structurally exposed. Ultimately, the fundamental challenge to a robust cybersecurity posture lies in institutional coherence rather than technical capability. Until these legislative deadlocks are resolved, even the most comprehensive attempts at public-private cooperation will be hindered by the very structural limitation.

3. Taiwan

3.1. PRC Cyber Operations and Hybrid Threats

Taiwan’s cybersecurity environment is increasingly shaped by persistent PRC cyber operations and broader hybrid threats. According to Taiwan’s National Security Bureau (NSB), cyberattacks originating from the PRC against Taiwan’s critical infrastructure—including hospitals, banks, energy systerns, and other key sectors—rose to an average of 2.63 million intrusion attempts per day in 2025, a 6 percent increase over the previous year (National Security Bureau of Taiwan 2025). Reuters reports that these operations reportedly intensified around PRC military exercises and Taiwan’s major political events, reinforcing Taiwan’s view that cyber operations form part of a broader strategy of hybrid coercion (Reuters 2026). Several incidents illustrate this threat. In January 2024, the semiconductor-related firm Foxsemicon was hit by a ransomware attack in which its website was defaced and the attackers threatened to publish customer and employee information (Wang 2024). In February 2025, Mackay Memorial Hospital in Taipei was also struck by ransomware, and later reports raised concerns over the possible leakage of approximately 16.6 million documents (Hiciano 2025). According to Taipei Times, later in April 2025, Taiwanese authorities identified Lo Chengyu as the suspect behind the attack, marking the first public identification of a Chinese hacker by Taiwanese law enforcement in such a case (Chung 2025).

Moreover, cyber operations in Taiwan are not treated merely as attempts to disrupt systerns or steal data; they are also linked to influence operations and disinformation. In its analysis of the PRC’s 2025 cognitive warfare tactics against Taiwan, NSB reported (National Security Bureau of Taiwan 2026) that during the April 2025 military exercises, PRC-linked actors hijacked more than ten PTT Bulletin Board systern accounts and used compromised IoT devices and rented overseas servers to spread false narratives, including that the PRC had blocked Taiwan’s natural gas shipments and that Chinese warships had entered Taiwan’s 24-nautical-mile zone[1]. Therefore, Taiwan’s cyber threat environment is understood as a compound security challenge in which infrastructure disruption and information manipulation are closely intertwined.

3.2. Cybersecurity in Taiwan’s Resilience Strategy

Against this background, cybersecurity has come to be positioned as a core component of Taiwan’s broader resilience strategy. The Whole-of-Society Defense Resilience Committee, announced by the Presidential Office in June 2024, was established to ensure that both government and society can continue to function during national emergencies or natural disasters (Office of the President of the Republic of China N.d.). In a September 2025 speech, President Lai Ching-te explicitly identified “cyber and financial security” as one of the committee’s five principal lines of effort. This prioritization is also reflected in Taiwan’s AI governance framework (Lai 2025). The AI Basic Act, which entered into force in January 2026 (Artificial Intelligence Basic Act 2026), sets out seven governance principles in Article 4, one of which is “cybersecurity and safety.” The Act requires that AI research, development, and deployment incorporate cybersecurity safeguards, prevent security threats and attacks, and ensure the robustness and safety of AI systerns.

3.3. Strengthening Taiwan’s Cybersecurity Law

At the legal level, the Cyber Security Management Act remains Taiwan’s core cybersecurity statute (Cyber Security Management Act 2025). It was enacted as Taiwan’s first cybersecurity-specific law and took effect on 1 January 2019 under the Executive Yuan. It applies not only to government agencies, but also to critical infrastructure providers, state-owned enterprises, and other designated non-governmental entities. In 2022, much of its supervisory authority was transferred to MODA.

A major shift came with the 2025 amendment, which elevated the regulation of high-risk digital products from adrninistrative practice to statutory law. Article 3(11) defines “products harmful to national cyber security” as information and communications systerns, services, or products that pose a direct or indirect risk to national cybersecurity and may affect government operations or social stability (Stellex Law Firm 2025). Article 11 prohibits government agencies from downloading, installing, or using such products, except in narrowly defined cases where no alternative exists and specific approval has been granted. Pursuant to Article 11(3), the Regulations for the Review of Products Harmful to National Cyber Security were issued in December 2025, further specifying review procedures, risk assessment, intelligence sharing, and use restrictions (Regulations for Review of Products Harmful to National Cyber Security 2025).

These legal changes were quickly connected to the regulation of Chinese digital products. Taipei Times reported that in December 2025, MODA publicly warned that Douyin[2], RedNote[3], Sina Weibo[4], Weixin[5], and Baidu Cloud[6] posed serious information-security risks and stated that government agencies were prohibited from downloading, installing, or using these applications on official devices and networks (Chiu and Chung 2025). Although MODA had already restricted the use of TikTok-related applications in the public sector by 2022 (Ministry of Digital Affairs of Taiwan 2022), the 2025 revision strengthened their legal basis and facilitated their broader implementation.■

[1]PTT (批踢踢實業坊). "PTT: Taiwan's Largest Online Forum." https://www.ptt.cc/bbs/index.html.

[2]Douyin (抖音). "Douyin: Chinese Short-Video Platform." https://www.douyin.com/jingxuan.

[3]Xiaohongshu (小紅書). "Xiaohongshu (RED): Chinese Social Commerce Platform." https://www.xiaohongshu.com/explore.

[4]Weibo (微博). "Weibo: Chinese Microblogging Platform." https://weibo.com/.

[5]WeChat (微信). "WeChat: Chinese Messaging and Social Media Platform." https://weixin.qq.com/.

[6]Baidu AI Cloud. "Baidu AI Cloud: Cloud and AI Services." https://login.bce.baidu.com/?lang=en.

Sovereign AI & Privacy

The governance of personal data and the pursuit of sovereign AI raise critical questions for democratic states regarding how to strengthen technological and economic capacity while safeguarding citizens’ rights and preventing the misuse or external exploitation of data. Japan has a relatively advanced and still evolving personal data protection framework, while GENIAC reflects a sovereign AI strategy framed primarily around industrial competitiveness. South Korea has taken a proactive approach to sovereign AI, exemplified by the Independent AI Foundation Model Project, but continues to face strong privacy and surveillance concerns, as highlighted by the iRuda case. Taiwan most clearly links open data to sovereign AI development, as seen in the Act for the Promotion of Data Innovation and Utilization and TAIDE, while its PDPA regime is being strengthened but remains institutionally incomplete.

1. Japan

1.1. Personal Data Protection Revised in the AI Era

Japan is currently advancing reforms to its personal data protection framework in response to the rapid expansion of generative AI. The central policy question is how to facilitate AI development while maintaining effective safeguards against the misuse of personal information. The recent policy document indicates that the government seeks to revise consent requirements under the Personal Information Protection Law while simultaneously strengthening sanctions against abusive practices (Personal Information Protection Commission of Japan 2026).

The core proposal would allow, under limited conditions, the acquisition and third-party provision of sensitive personal data without individual consent when the purpose is restricted to statistical processing related to AI development. According to press reporting (Kawashima and Murai 2025), this change reflects industry concerns that existing consent rules impede large-scale data collection necessary for AI training. At the same time, the draft includes the introduction of a surcharge systern targeting operators who intentionally misuse personal data for profit, additional safeguards for minors under 16, and procedural requirements for the handling of facial feature data. Although a collective redress mechanism was discussed during deliberations, it was not incorporated into the current proposal.

The appropriate use of personal data is also connected to a trustworthy democratic society. The 2025 AI Business Guidelines articulate expectations for responsible data practices (Office of National Security 2024). The Guidelines refer to overseas cases in which personal data collected through social media personality-quiz applications were used to classify individuals into behavioural groups and deliver targeted political messaging.

1.2. Sovereign AI under Development

In addition to personal data governance, Japan has also begun to pursue elements of a sovereign AI strategy. METI and the New Energy and Industrial Technology Development Organization (NEDO) have promoted the Generative AI Accelerator Challenge (GENIAC)[1] to support the development of domestic foundation models and their social implementation. The AI Basic Plan similarly emphasizes strengthening national AI development capacity as one of its core policy directions (Cabinet Office of Japan 2025). Materials (Cabinet Office and Ministry of Economy, Trade and Industry of Japan 2026) in the working groups in the Growth Strategy (Ministry of Internal Affairs and Communications 2026) show that the government seeks to build a domestic AI ecosystern, including data, data centers, foundation models, and applications, in order to reduce reliance on foreign technologies and to ensure that AI systerns reflect Japanese language, knowledge, and social contexts. However, these discussions are framed primarily in terms of industrial competitiveness and technological capability rather than personal data protection.

2. South Korea

2.1. Data Privacy Frontline

The AI privacy landscape is defined by the tension between data-driven innovation and information rights. The “iRUDA" case (Beopryul Shinmun 2021) remains a symbolic early instance of public distrust in AI data ethics. Currently, tech firms, including startups and the financial sector, argue that the traditional "consent" principle is a critical barrier to LLM development (Choi 2025), as individual consent for billions of data points is practically impossible. While the government’s "2025 AI Regulatory Rationalization Roadmap" seeks to mitigate legal risks for pseudonymized data (Government of the Republic of Korea 2025), civil society groups like People’s Solidarity for Participatory Democracy demand strict accountability for unauthorized data usage, ensuring innovation does not erode fundamental privacy (Civil Society Organizations 2021).

2.2. Sovereign AI and the Dilemma of Public AI Utilization

To secure data sovereignty, Korea strategically empowers domestic tech leaders and startups with computing resources like hardware and massive datasets to build a Sovereign AI ecosystern (Ministry of Science and ICT 2025b) independent of global Big Tech. Notably, under the Lee adrninistration, the development of sovereign AI has elevated it as a core pillar of the national AI strategy (Ministry of Science and ICT 2025c), illustrated in the Independent AI Foundation Model Project (Ministry of Science and ICT 2026a). Simultaneously, public-sector AI is expanding physically such as smart cities that utilize AI-integrated CCTV networks for automated threat detection (Ministry of Land, Infrastructure and Transport 2024). While this raises significant surveillance concerns, in a move for democratic values, the government accepted the National Human Rights Commission of Korea (NHRCK)’s advice (National Human Rights Commission of Korea 2024), banning real-time facial recognition in public agencies pending legislative action. This reflects a mature societal consensus that human rights and privacy must effectively counterbalance technological efficiency.

Korea’s high public tolerance for data-driven governance, reinforced by the large-scale mobilization of CCTV, credit card, and mobile data during the COVID-era, has significantly lowered societal friction for comprehensive data collection. On the contrary, this trend exists alongside deep structural tensions, where the “iRUDA" fallout and consistent civil society pushback reflect a landscape where data ethics remain genuinely contested. While the tentative ban on real-time facial recognition signals a degree of institutional self-restraint, the simultaneous expansion of smart cities suggests that such restraint is often selective and pragmatic rather than principled. Consequently, Korea sits at a crossroads where adrninistrative pragmatism and rights-based advocacy coexist uneasily.

3. Taiwan

3.1. Data Utilization and Sovereign AI

Taiwan’s data governance strategy has increasingly linked data circulation to the development of sovereign AI. In 2025, MODA published the draft Act for the Promotion of Data Innovation and Utilization (Ministry of Digital Affairs 2025), which seeks to institutionalize data flows in the AI era by improving open-data quality, promoting routine inter-agency and industry data sharing, standardizing licensing terms, and introducing Chief Data Officers across government. The draft also envisages an Executive Yuan Data Innovation and Utilization Advisory Council to coordinate data policy across ministries at the cabinet level.

These measures are significant not only because they facilitate data reuse, but also because they help build the training infrastructure needed for AI systerns that reflect Taiwan’s linguistic, institutional, and social context. In this sense, expanded data utilization is closely tied to Taiwan’s broader sovereign AI agenda, which aims to reduce excessive reliance on foreign foundation models and non-local language data. This, in turn, may create greater room for Taiwan to govern sensitive industrial data (Murphy 2024) and citizen data in accordance with domestic institutions and values. At the center of this agenda is the Trustworthy AI Dialogue Engine (TAIDE)[2], which provides five core functions (Taiwan Panorama 2026)—Chinese-English and English-Chinese translation, email drafting, summarization, and writing assistance—and is explicitly designed to strengthen understanding of Taiwanese language and culture. In addition, MODA announced in December 2025 the launch of the Taiwan Sovereign AI Training Corpus (Ministry of Digital Affairs 2025), which aggregates more than 2,000 datasets and over 600 million tokens from more than 200 government agencies, thereby furnishing a more robust training base for sovereign AI development.

3.2. Privacy and Surveillance Concerns

At the same time, however, the expansion of data utilization and sovereign AI has generated growing tension with personal data protection. In Taiwan, a 2023 amendment to the Personal Data Protection Act (PDPA) created the legal basis for an Independent Personal Data Protection Commission (PDPC) to become the competent authority (Legislative Yuan 2025), and in December 2023 the government established the Preparatory Office of the PDPC[3] to lay the groundwork for its future operation. A further partial amendment to the PDPA (Chang and Hsiung 2025), submitted by the Executive Yuan in March 2025 and passed by the Legislative Yuan in November 2025, strengthened the rules on breach notification, response measures, and record retention. Nevertheless, as of November 2025, the Preparatory Office of the PDPC (Personal Data Protection Commission Preparatory Office 2025) had still not been formally established, because the legislation required for its establishment had not yet completed the legislative process.

Yet, while these developments reflect Taiwan’s effort to strengthen privacy protection at the institutional level, they have not eliminated emerging risks associated with new forms of data use. Privacy concerns have also surfaced in the political sphere. According to TFC, controversy erupted in late May 2024 after DPP politician Wang Yi-chuan suggested that the ages and other characteristics of protesters near the Legislative Yuan could be inferred through mobile signaling data, prompting criticism that such analysis could amount to political surveillance. The ensuing debate centered on the legal basis for accessing such data, as well as on the effectiveness of consent and anonymization safeguards.■

[1]Ministry of Economy, Trade and Industry (METI), Japan. "GENIAC: Generative AI Accelerator Challenge." https://www.meti.go.jp/policy/mono_info_service/geniac/index.html.

[2]TAIDE. "TAIDE: Trustworthy AI Dialogue Engine, Taiwan." https://taide.tw/en.

[3]Personal Data Protection Commission (PDPC), Taiwan. https://www.pdpc.gov.tw/en/.

Copyright & Intellectual Property Rights

Approaches to generative AI and copyright differ primarily in the degree of legal certainty surrounding AI training data. Japan provides the clearest framework through Article 30 (4) of the Copyright Act, which broadly permits the use of copyrighted works for machine learning, although debates over intellectual property and creators’ rights remain ongoing. South Korea faces a regulatory gap, and conflicts between platforms and creators have been surfacing. In Taiwan, the legality of AI training remains uncertain and is assessed under fair use. Together, these cases illustrate the broader tension between promoting AI innovation and protecting the informational foundations of democratic societies.

1. Japan

1.1. Permissive Environment for AI Training

Japan’s approach to generative AI and copyright combines a relatively permissive legal framework for AI training with ongoing debates over the protection of creative industries and media outlets. The current policy discussion struggles to balance innovation in AI development with safeguarding authors’ rights.

A distinctive feature of the Japanese systern is the copyright exception for information analysis under Article 30-4 of the Copyright Act, which allows copyrighted works to be used without permission when processed for information analysis, including machine learning. Due to the broad scope of this exception, some have called Japan a “machine learning paradise.” (Ueno 2021) This provision, introduced in 2018 to promote technological progress, is seen by by some as inappropriate in the AI era (Takabe 2025).

However, the rapid spread of generative AI has raised societal concerns about potential copyright and intellectual property infringement. In response, the Agency for Cultural Affairs issued guidance in 2023 (Japan Copyright Office 2024). It clarifies that, while AI training continues to fall under the information-analysis exception, AI outputs could still infringe copyright if, for example, a systern is designed to produce works similar to those used in training. Nonetheless, the legal systern remains favorable toward the interests of AI developers.

1.2. Ongoing Debates on the Rights-Protective Proposal on Intellectual Rights

However, the newly proposed guideline on property rights adopts a more rights-based approach and has become a subject of controversy. In 2025, the government proposed a draft of the principal code (Ueno 2024) framework for AI developers, which appears to be quite strict. Instead of statutory obligations, the code encourages a compliance-or-explain approach that urges companies to respect rights holders’ interests. The draft requires AI developers and providers to disclose overviews of their models and training data, provide additional information if requested, and explain non-compliance with the code.

Different stakeholders hold varying views. Industry groups (Japan Association of New Economy 2026) worry that stricter rules could limit AI growth, while newspaper publishers (Japan Newspaper Publishers & Editors Association 2026) and creator organizations (NAFCA 2026) generally support stronger copyright protections. The publishers even call for enhanced safeguards. Meanwhile, some believe the draft is unrealistic for AI developers in the first place and question the efficacy of the domestic approach, as the issue is transnational (Nikkei 2026a). How the policy debate will resolve remains uncertain.

With technologies like Sora2 enabling anyone to produce high-quality content, (1) the creative industry (Japan Animation Association et al. 2025) warns that generative AI could threaten employment for artists and media professionals if AI systerns are trained on existing works without permission, transparency, or compensation. These concerns can extend beyond copyright, as a decline in professional creative work might undermine the cultural and informational fabric essential to democratic societies.

On the other hand, AI technology may protect creators’ rights. It has been reported that Sony AI has been developing a systern that detects copyrights infringement by LLMs and ensures fair compensation for creators (Nikkei 2026b).

2. South Korea

2.1. Legislative Shortcomings and the Widening Data Divide

The South Korean government maintains a permissive stance toward AI development through regulatory rationalization. However, the National Assembly Research Service has criticized the inadequate management of corresponding copyright frameworks (Baek 2025). This legislative vacuum has offered incentives on companies to shift toward contract-based data usage to mitigate legal risks (National Assembly Research Service 2025). Concerns are mounting that this trend significantly widens the gap in access to high-quality data between large corporations with superior capital and resource-lacking companies. Such a disparity risks solidifying a data monopoly centered around mega platforms.

2.2. Structural Conflicts between Platforms and Creators

Structural conflicts are intensifying between platforms seeking to utilize content for AI training and creators striving to protect their intellectual property. For instance, platforms like Naver have faced backlash (Lim 2023) for using hosted works to provide AI services that replicate specific artistic styles provided by authors distributing on their platforms. On the other hand, groups such as the Korea Webtoon Author Association have condemned these practices (KBS News 2023), arguing that they push creators, who are placed in subordinate contractual positions, into a corner. This friction highlights a power asymmetry where platforms leverage their distribution dominance, underscoring the urgent need for a framework that balances technological utility with the rights of creators.

Beyond just creating uncertainty, Korea’s current copyright vacuum acts as a structural accelerator. It quietly advantages those with existing capital, potentially turning AI into a new dimension where the gap between large corporations and smaller players, in talent and infrastructure, widens further. This disparity is rooted in a profound structural power asymmetry, where the dominant distributing platforms assume actual superiority over intellectual property rights, leaving subordinate creators to bear a disproportionate share of AI’s negative influence. Ultimately, Korea’s innovation-first approach is creating real structural casualties, signaling that sustainable AI development will require frameworks that go beyond corporate risk mitigation, including actively broadening data access and giving creators meaningful legal protection.

3. Taiwan

3.1. Legal and Institutional Framework

Taiwan’s legal framework governing AI and copyright is currently shaped primarily by interpretations of the existing Copyright Act (Legislative Yuan 2022), rather than by AI-specific statutory provisions. The Act grants copyright holders’ exclusive rights such as the right of reproduction (Article 22), the right of public transmission (Article 26 (1)), and the rights of adaptation and compilation (Article 28). Accordingly, copyright may be implicated at several stages of AI development, including data collection and reproduction, the use of copyrighted works for training, and the distribution or adaptation of AI-generated outputs.

With regard to the copyrightability of AI-generated works, the Taiwan Intellectual Property Office (TIPO) Email No.1140522c (Intellectual Property Office 2025) stated that where a work is generated entirely through AI computation, without human intellectual or creative contribution, it does not qualify for copyright protection. At the same time, TIPO Email No.1111031 explained that where AI is used only as an assistive tool and the creator contributes creative input to the final work, copyright protection may still apply.

The legal status of AI training, however, remains much more uncertain. As Yasuto Shirae notes (Shirae 2025), Taiwan does not have an express exception comparable to Article 30 (4) of Japan’s Copyright Act for information analysis or AI training (National Diet 1970). Instead, the legality of AI training is assessed case by case under the Fair Use provision of Article 65 of the Copyright Act. Article 65(2) requires consideration of four factors: the purpose and nature of the use, the nature of the copyrighted work, the amount and substantiality of the portion used, and the effect of the use on the work’s market value. As of March 2026, no publicly available Taiwanese court decision appears to have definitively resolved whether the use of copyrighted works for generative AI training qualifies as fair use under Article 65 of the Copyright Act.

3.2. Illustrative Cases

One illustrative case concerns TAIDE and the management of copyright-related legal risk. This legal uncertainty is reflected in the development of Taiwan’s sovereign AI project, TAIDE[1]. Yasuto Shirae argues that the training of large language models such as TAIDE cannot safely be assumed to fall within fair use and may therefore require authorization from rights holders (Shirae 2025). TAIDE’s own disclosures suggest an awareness of this risk. The TAIDE website[2] lists numerous government and private-sector institutions as cooperating data providers, and some pages on its training data[3] explicitly state that the materials were provided under authorization. In addition, the TAIDE team stated that the TAIDE-LX-7B model was trained on text data lawfully licensed through consultations with government agencies and private publishers (TAIDE 2024). TAIDE can thus be understood as a case in which legal uncertainty surrounding fair use is managed through licensing and data provenance control.

A second illustrative case can be found in disputes over the use of news content as training data. These tensions are particularly visible in the news industry. The Central News Agency (CNA)[4] stated that approximately 140,000 of its news articles published between 2011 and 2021 had been included without authorization in the Traditional Chinese dataset fineweb-zhtw (Lin et al. 2024). CNA requested the removal of the materials and filed a criminal complaint; it argued that news should not be treated merely as a resource for extraction, but as a public good deserving respect and protection. According to CNA[5], the dispute was resolved in July 2025 after the opposing party acknowledged the copyright infringement and deleted the relevant materials. More broadly, however, the case highlights the absence of clear legal rules in Taiwan governing the use of news content for generative AI training, particularly with regard to authorization, compensation, and copyright clearance.■

[1]TAIDE. "TAIDE: Trustworthy AI Dialogue Engine, Taiwan." https://taide.tw/en.

[2]TAIDE. "TAIDE Training Data." https://taide.tw/public/training-data.

[3]TAIDE. "TAIDE Training Data Details." https://taide.tw/public/training-data_show?id=4b1141869c45c11c019c466de42c0011.

[4]Central News Agency (CNA) Taiwan. July 7, 2025. Facebook post. https://www.facebook.com/cnanewstaiwan/posts/1127939982697899.

[5]Central News Agency (CNA) Taiwan. July 11, 2025. Facebook post. https://www.facebook.com/cnanewstaiwan/posts/1131023115722919.