Rudolf BerkesAt a September 18, 2025 event, Hungarian officials and legal experts convened to discuss the national implementation of the European Union's landmark AI Act. The discussion, hosted by the Institute for Legal Studies, National Laboratory for Artificial Intelligence and the Algorithmic Constitutionality Momentum Research Group, featured Deputy State Secretary for Technology Szabolcs Szolnoki and Research Professor Zsolt Ződi, who provided critical insights into the planned regulatory structure and the anticipated hurdles ahead. Supplemented by draft legislation, their comments paint a picture of a centralized, consultative approach that relies heavily on soft law to navigate the complexities of the new regulation. The event was moderated by Senior Research Fellow Kitti Mezei.
The Institute for Legal Studies has launched a recurring interdisciplinary forum to address how artificial intelligence is reshaping legal work. The series brings together legal scholars, practitioners, social scientists, IT experts, and professionals from state, business, and civil sectors to explore both the advantages and challenges of AI in law, especially in light of the EU’s AI Act implementation. Discussions focus on long-term, sustainable solutions by considering technological, legal, and social science perspectives, with topics such as algorithmic transparency, reliability of legal research, and intellectual property in creative AI outputs. Held every two months, these dialogues aim to reduce legal uncertainty, identify synergies across disciplines, and prepare legal professionals for AI’s integration into their everyday practice. The first of such event on September 18, covered the implementation of the EU AI Act in Hungary.
The cornerstone of Hungary's implementation will be a two-tiered institutional system: a central AI Office to handle administrative and market functions, and a broader AI Council to provide strategic oversight and professional guidance.
The Hungarian AI Office (MI Hivatal) is designed as a "one-stop-shop" to simplify compliance for businesses. It will consolidate the market surveillance authority, which will be an internal unit of the Ministry for National Economy, and the single point of contact, both roles designated to the minister responsible for business development. The crucial function of the notifying authority—which designates and monitors the conformity assessment bodies that audit high-risk AI—will be handled by the National Accreditation Authority.
Complementing this office is the Hungarian AI Council (MI Tanács), envisioned as the domestic counterpart to the EU AI Board. Szabolcs Szolnoki noted that the council's composition was expanded following public consultation from an initial concept of a small group of authority leaders to a much broader body. As outlined in the draft legislation, its membership will include key authorities such as the National Bank, the Competition Authority, and the Data Protection Authority, alongside ministries and academic institutions like the Hungarian Academy of Sciences. The Council will have a wide-ranging mandate, from shaping the national AI strategy and advising the government to coordinating research and playing a key role in designing the regulatory sandbox.
During the event, the speakers delved into the practical and philosophical difficulties of applying the AI Act. Research Professor Zsolt Ződi addressed the fundamental issue of defining an AI system. While he initially found the Commission's "soft, philosophical" definitions impractical for IT professionals, he concluded that most organizations will ultimately be able to identify which of their systems fall under the Act's purview. More significantly, he stressed that in the public sector, nearly all AI systems are likely to be classified as high-risk.
A major concern raised by Ződi was the Act's horizontal nature, attempting to regulate vastly different sectors with a single framework. He expressed doubt about this approach and highlighted the immense challenge of harmonizing the AI Act with existing digital laws, such as data protection. "In EU digital lawmaking," he argued, "we don't think through the relationship between the rules," creating a complex ecosystem where even small changes are difficult to implement. He further criticized the tendency of modern governments to over-regulate in an attempt to fulfill an "unfulfillable promise" to protect citizens from all harm, often without a full understanding of the societal impacts.
Ződi was particularly skeptical about the enforcement of the Act's human rights impact assessment requirements for high-risk systems. He voiced strong doubts, predicting it would become a mere "ticking exercise" for auditors rather than a genuine impact assessment. He explained that human rights impacts are difficult to quantify and do not fit well with a product-compliance approach, despite the noble intentions of the lawmaker.
Both speakers agreed on the critical importance of soft law. Szabolcs Szolnoki stated the government's intention is to regulate only the mandatory elements and to "use soft law for most things" to avoid creating a convoluted system. Zsolt Ződi went further, asserting that the main focus of compliance will inevitably be on soft law because a purely hard-law-centered approach is unworkable.
On the topic of transparency, another cornerstone of EU digital regulation, Ződi offered a cautionary note. While foundational models will be subject to transparency rules, he warned that this can create a "sea of data" that is impossible for the public to analyze. He emphasized the need for experts, researchers, and civil society to interpret this data and make it understandable.
The Hungarian implementation plan includes all mandatory elements of the EU AI Act, such as the establishment of a regulatory sandbox by August 2026 to allow for the safe testing of innovative AI. Regarding sanctions, Szabolcs Szolnoki, who serves as Hungary's national delegate to the EU AI Board, shared that while there is openness among member states, there is also "a secrecy due to the competitive situation," particularly concerning the design of sanctioning systems. In Hungary, the AI Council will play a significant role in developing these penalties. The draft government decree sets the maximum fines in forint, with the highest penalty for prohibited AI practices reaching 13.3 billion HUF, in line with the EU Act's tiers.
Looking ahead, Szolnoki affirmed the government's commitment to continued dialogue, promising to bring feedback from professional and academic circles to future meetings of the EU AI Board. He invited the legal scholars to participate in public consultation and help policymakers better understand the complexities and far-reaching consequences of AI regulation. Ződi believes that legal scholars will need to systematize and scrutinize the constantly evolving EU digital regulation landscape, while adapting traditional approaches to the new policy-based compliance requirements.
The next event in the series will take place on November 27.
__________________________________________________________
This report was prepared with the support of the Algorithmic Constitutionalism Research Group (LP2024-20/2024), funded by the Hungarian Academy of Sciences.
__________________________________________________________
The views expressed above belong to the author and do not necessarily represent the views of the Centre for Social Sciences.