Legal AI agents in EU and South Korea

By Efstathios Iliopoulos and Somi Jeong

AI agents have emerged as a highly debated development across various industries. In this essay, we try to shed light on some key aspects regarding the use of agents in the legal industry, through the comparison between the EU and South Korean modus vivendi. It demonstrates that the effectiveness of AI agents in the legal sector is shaped less by regulatory caution alone and more by underlying digital infrastructure, with the European Union’s fragmented ecosystem limiting deployment, in contrast to South Korea’s integrated and data-driven framework, which actively enables it.

An AI agent is a computational system, mostly built on large language models, that is capable of perceiving inputs from a given environment, making decisions through reasoning and planning, and executing actions to achieve a defined goal. Unlike standard machine learning models that generate single-step outputs, AI agents operate in a continuous loop of observation, reasoning, and action, often incorporating tools, memory, and feedback mechanisms to complete multi-step tasks autonomously[1].

In the legal industry, AI agents are not just theoretical, they are already being implemented in several practical ways. For example, firms use them for contract analysis, where an agent reviews agreements, identifies clauses, flags inconsistencies, and even suggests edits. In legal research, AI agents can search case law, summarise judgments, and extract relevant precedents far faster than manual methods. They are also used in document automation, drafting standard contracts or filings based on user inputs, and in compliance monitoring, where they track regulatory changes and assess organisational risk exposure. More advanced implementations include agents that can assist with litigation strategy by analysing large datasets of past cases or even simulate negotiation scenarios.

However, their deployment comes with important limitations and risks. Legal work requires high accuracy, accountability, and confidentiality and AI agents can still produce errors (“hallucinations”[2]) or miss nuanced legal interpretations. There are also ethical and regulatory concerns, especially under frameworks like the EU AI Act, which classifies many legal AI applications as high-risk. As a result, AI agents in law are typically used, especially in the EU, as decision-support tools rather than autonomous decision-makers, augmenting lawyers rather than replacing them.

But is the risk of hallucinations or the restrictions imposed by the EU AI Act or national legislation or institutional codes the only reason for the limited approach of agents in EU so far?

The agent, in order to perform, requires well-organised data and a digitally friendly judicial and public environment. Let’s examine what is the case in both the EU and the Korean market, to pick up similarities and differences between them.

The digitalisation of justice systems across the European Union has advanced significantly, yet remains uneven and fragmented. According to the European Commission’s EU Justice Scoreboard 2025[3], most Member States now provide at least some level of online access to justice, including the possibility to initiate proceedings electronically, consult case information and communicate with courts and their administrations via digital channels. The Courts themselves rely more and more on case management systems, electronic files and digital registers, reflecting a gradual shift away from paper-based processes. The expansion of videoconferencing, particularly since the COVID-19 period, has further accelerated the adoption of remote hearings in many jurisdictions. These developments indicate that the foundational elements of digital justice are now widely established across the EU.

However, the Scoreboard also highlights persistent limitations that prevent the emergence of fully digital justice systems. The availability and scope of digital tools vary considerably between Member States, as well as across different types of proceedings within the same country. In many cases, electronic communication is restricted to legal professionals, while full online access for citizens remains limited. The use of digital evidence and fully digital procedures is still underdeveloped, with only a small number of countries offering comprehensive solutions across all case types. Moreover, cross-border interoperability of digital systems remains fragmented, hindering the development of a seamless European judicial area. Overall, the findings suggest that while digitalisation is progressing, EU justice systems are still in a transitional phase characterised by partial integration rather than full digital transformation.

[4]

But this scoreboard doesn’t reveal the whole picture. If we want a more holistic view, we should take into account the  “Public services for citizens” indicator, used in the European Commission’s digital government monitoring frameworks such as the Digital Economy and Society Index[5] and the eGovernment Benchmark, measures the extent to which key administrative services are available and usable online for individuals. It captures both the breadth and sophistication of digital public services, assessing whether citizens can complete procedures—such as applying for documents, registering residence, or accessing social benefits—entirely online without the need for physical presence. The indicator typically combines elements such as service availability, online completion rates, usability, and the degree to which services are designed around user needs (user-centricity). As such, the “Public services for citizens” indicator serves as a key proxy for the maturity and accessibility of eGovernment systems, highlighting both progress in digital transformation and remaining gaps in inclusiveness, cross-border access, and end-to-end digital service delivery across EU Member States.

The results of the “Public services for citizens” indicator show that digital public service provision across the EU has reached a relatively advanced stage (above the EU average level), but remains uneven between Member States. According to the Digital Economy and Society Index and the eGovernment Benchmark, a majority of countries now offer a wide range of administrative procedures fully online, with particularly high performance observed in Northern and Baltic countries such as Estonia, Denmark, and Finland. These countries achieve near-complete digital service delivery, supported by strong digital infrastructure and widespread use of electronic identification systems. In contrast, some Eastern European countries still lag behind, often due to limited service availability, lower levels of usability, or incomplete end-to-end digitalisation.

Overall, the EU average indicates steady improvement over time, especially in terms of service availability and user-centric design. However, important gaps persist in areas such as cross-border accessibility, inclusiveness for digitally vulnerable groups, and the full integration of back-office systems. In many cases, services are only partially digitalised, requiring users to complete certain steps offline or interact with multiple administrative bodies. The findings therefore suggest that, while the EU has made substantial progress in digitising public services for citizens, further efforts are needed to achieve fully seamless, interoperable, and universally accessible eGovernment systems across all Member States.

This table, albeit its optimism, should be read better in conjunction with the pre-filled forms indicator, (used in the same frameworks), that measures the extent to which public administrations reuse existing data to automatically populate online forms, thereby reducing the need for citizens to repeatedly enter the same information. It serves as a key proxy for the level of back-office integration and the implementation of the “once-only” principle in digital government. Results across the EU reveal a clear divide: digitally advanced countries, particularly in Northern and Baltic Europe, achieve high scores due to strong interoperability frameworks and effective data-sharing practices, while many other Member States lag behind because of fragmented administrative systems and legal or technical barriers to data exchange. Consequently, even where online services are available, users are often still required to manually input information, highlighting that the main challenge for the EU lies not in the availability of digital services, but in the deeper integration of public sector data systems. This is mainly due to fragmented administrative structures, strong data protection constraints, and relatively low levels of interoperability between public sector databases across federal, regional, and local levels.

The quantitative indicators discussed above do not fully capture how digitalisation operates in everyday judicial practice. While the EU is generally based on civil law systems (continental law) and places strong emphasis on procedural guarantees, fundamental rights, and data protection standards, the actual implementation of digital tools in legal proceedings remains uneven across Member States. In practice, many judicial systems are still in a hybrid phase, where digital procedures coexist with paper-based workflows, particularly in civil and administrative cases. This reflects not only technological capacity but also institutional fragmentation, varying administrative structures, and differing levels of interoperability between courts and public authorities. As a result, despite significant progress in e-justice initiatives, the digital transformation of EU judicial systems remains incomplete and continues to evolve gradually rather than uniformly across the Union.

The situation in South Korea presents a contrasting model.

South Korea’s legal system, shaped in the early twentieth century by German and Japanese legal frameworks, shares the civil law tradition with most of the EU and still looks to codified statutes first. Precedents are treated as reference materials rather than binding principles. Undeniably, a strict comparison between EU and Korea is not feasible, since the EU comprises 27 Member States. Nevertheless, some general assumptions can be implied.

South Korea has obviously operated a single judicial and administrative system under one jurisdiction. That structural unity is what allowed both conditions to be met quickly and consistently, and it is where the comparison with the EU becomes instructive.

South Korea’s judicial infrastructure has reached a fairly mature stage of digitalisation. Electronic litigation is nearly universal in civil and administrative proceedings. A single integrated system under the Supreme Court centralises filing, service, and record management. At the executive level, the Ministry of Government Legislation provides statutes, precedents, local ordinances, treaties, and administrative rules as an official Open API through its National Legislation Information Sharing Service[6]. Researchers, developers, and ordinary citizens alike can obtain a free API key with basic identity verification. The government’s public data portal[7] further opens non-legal administrative datasets through standardised APIs.

This infrastructure goes well beyond simply hosting legal documents on the web. Notably, the State provides legal information in a structured, machine-readable form. Where EU judicial digital tools remain uneven across Member States and procedure types (still in a transitional phase where paper and digital coexist), South Korea’s legal and judicial data goes beyond formal openness: it is available for immediate use on top of an official, national-level API layer. South Korea already substantially meets the first condition.

The second condition is embedded in the Korean legal system itself. The civil law tradition explicitly structures legal principles in the form of written statutory provisions. Unlike common law systems built on precedent, this framework is inherently closer to an explicit rules system. The ‘simplicity’ here is not the result of deliberate simplification; it is a characteristic intrinsic to the legal system from the outset. How this structural feature is being leveraged in actual AI agent development can be observed concretely in Korean legal AI open-source projects released over the past two to three years.

South Korea already has legal AI projects leveraging these structural conditions, published in open-source form. Notable examples include an automated privacy policy generator based on the Personal Information Protection Act (PIPA)[8], a contract drafting tool incorporating the latest Supreme Court precedents, and a server that wraps 41 official legislation APIs from the Ministry of Government Legislation into 16 Model Context Protocol (MCP) tools[9]. The last project includes a built-in citation verification mechanism designed to block hallucinations, a technical response made possible precisely because the statutory data is structured. These tools are all built on international AI agent standards such as Anthropic’s Model Context Protocol and are updated promptly in line with legislative amendments. This signals that South Korea’s legal AI ecosystem operates in conjunction with open international standards.

The contributors behind these projects fall clearly into two streams. The earlier tools are led by private entrepreneurs working in the legal technology space. The most widely adopted of them (the legislation API-based MCP server) is built and maintained in a personal capacity by a serving local government official. This structure, where a private entrepreneur and a public sector insider contribute simultaneously on the same open-source layer, characterises how South Korea’s legal AI ecosystem has taken shape.

The MCP server case illustrates well what becomes possible for an individual practitioner when three layers converge. The developer in question is a serving local government official with an economics background and no formal software engineering training, who describes this project on his GitHub profile as a ‘side project’. He has publicly stated that he built it through iterative refinement of AI-generated code, an approach now known as “vibe coding[10]”. Structured legislative data provided through an official Open API, an explicit rules system offered by codified law, and generative AI lowering the barrier to coding: when these three layers align, a domain practitioner can build the AI agent infrastructure for their own field themselves.

This capability, however, also reveals a structural gap: the infrastructure comes from individuals rather than from the State, meaning that the technical conditions are already in place. The Korean case, moreover, does not simply validate the broader logic. Despite significant progress in digitalisation and codification, the application of AI agents to legal practice remains uneven across task types. In structured document work (standard contracts, regulation-based compliance documents, formatted administrative outputs), clear results are observed; by contrast, in cases involving ambiguous legal interpretation, contextual assessment of local precedents, or multi-jurisdictional complexity, human legal judgement remains decisive. Gaps persist on the adoption side as well: large law firms are integrating both international enterprise solutions and domestic tools, public institutions tend to favour domestic options, the domestic SME legal market remains largely without appropriate tools, and document format interoperability is not yet a fully resolved challenge.

The implication of the South Korean case is nonetheless clear. The proposition that AI agents perform more effectively when a legal system is digitalised and structured in a digitally friendly way is not a merely theoretical claim; it is a phenomenon that can actually be observed in an environment where those conditions are met. At the same time, the Korean path leaves the EU with two takeaways. First, digitalisation and codification are necessary conditions, but not sufficient ones. The institutional coherence (a single jurisdiction, a single language, and centralised data standards) significantly amplifies the effect of those conditions. This is why the debate around legal AI agents in the EU may need to expand beyond technology and regulation to encompass the question of institutional infrastructure.

Therefore, the European Union should proceed simultaneously on two fronts: at the national level, by advancing the digitalisation and structuring of judicial and administrative systems, and at the EU level, by strengthening interoperability, common standards, and cross-border data integration. Only through this dual approach can the EU create the conditions necessary for the meaningful and scalable deployment of AI agents in the legal sector.

[1] RSC Publishing (2025) Autonomous AI agents and reasoning systems. Available at: https://pubs.rsc.org/en/content/articlehtml/2025/sc/d4sc03921a (Accessed: 22 April 2026).
AIWiki (n.d.) AI agent. Available at: https://aiwiki.ai/wiki/ai_agent (Accessed: 22 April 2026).

[2] See Efstathios Iliopoulos’ article of the heckelai.com April 2026 edition.

[3] https://commission.europa.eu/strategy-and-policy/policies/justice-and-fundamental-rights/upholding-rule-law/eu-justice-scoreboard_en

[4] Table based on a synthesis of qualitative assessments from the EU Justice Scoreboard indicators.

[5] https://digital-decade-desi.digital-strategy.ec.europa.eu/datasets/desi/charts/compare-countries-progress?indicator=desi_pff&breakdown=all_egov_le&unit=egov_score&country=BG,DE,EL,HR,DK

[6] Ministry of Government Legislation (n.d.) National Legislation Information Sharing Service. Available at: https://open.law.go.kr/LSO/openApi/guideResult.do (Accessed: 23 April 2026).

[7] Ministry of the Interior and Safety (n.d.) Public Data Portal of Korea. Available at: https://www.data.go.kr (Accessed: 23 April 2026).

[8] Personal Information Protection Act of Korea (Act No. 10465, 2011; most recently amended 2023). Personal Information Protection Commission. Available at: https://www.pipc.go.kr/eng/user/lgp/law/lawsRegulations.do (Accessed: 23 April 2026).

[9] Selected Korean legal AI open-source projects. Available at: https://github.com/chrisryugj/korean-law-mcp ; https://github.com/kimlawtech/korean-privacy-terms ; https://github.com/kimlawtech/korean-contracts (Accessed: 23 April 2026).

[10] Vibe coding is a software development trend where applications are built primarily through natural language conversation with AI, rather than manually writing lines of code. The term was coined by AI researcher Andrej Karpathy (formerly of OpenAI and Tesla) in early 2025.