Ádám EszterWith the professional support of the National Laboratory for Artificial Intelligence and the Momentum/Lendület Research Group on Algorithmic Constitutionalism, the ELTE CSS Institute for Legal Studies launched its scientific event series in the autumn of 2025 titled "The Sustainable Use of Artificial Intelligence in Legal Work." The fourth event of the series took place on April 9, 2026, focusing on the complex theoretical and practical issues regarding the judicial application of artificial intelligence. During the roundtable discussion, Dr. Adrienn Nagy (Associate Professor and Head of Institute, University of Miskolc) and Dóra Pálfi (PhD student, Károli Gáspár University of the Reformed Church) discussed the impacts of technological innovation on the administration of justice, moderated by Dr. Boldizsár Szentgáli-Tóth, the principal investigator the Momentum/Lendület Research Group on Algorithmic Constitutionalism. The objective of this analysis is to synthesize the key findings of the expert panel, with a particular focus on evidentiary challenges, the doctrine of legal liability, and the future development of the regulatory environment.
The starting point of the discourse was the phenomenon of "deepfake" content produced by generative artificial intelligence, which fundamentally undermines trust in the authenticity of court evidence. Experts agreed that technological advancement has reached a critical juncture where manipulated video and audio recordings are indistinguishable from reality to human senses. Dóra Pálfi argued that the limitations of technical defences, such as software detection, are evident, as the development of forging algorithms almost immediately follows and bypasses recognition systems. Despite this, the professional consensus suggests that fundamental changes to the Code of Civil Procedure or the introduction of specific deepfake regulations are unnecessary. Existing procedural frameworks – particularly the judicial right to the free evaluation of evidence – provide sufficient latitude to handle such abuses. The challenge lies more in increasing awareness in judicial practice and tightening source criticism.
One direct consequence of technological uncertainty may be an unexpected appreciation for the role of traditional means of evidence, particularly witness testimony and personal presence. Adrienn Nagy pointed out that due to the questionable integrity of digital evidence, the principle of immediacy becomes the primary guarantee of legal certainty. Procedural difficulties of a "trial within a trial" nature – where the court must decide on the technical authenticity of evidence rather than the merits of the legal dispute – could significantly increase the cost and duration of proceedings. While expert evidence has thus far been able to keep pace with technology, the threat of "live deepfakes" (real-time video manipulation) suggests that the physical courtroom and personal presence may remain the only environments where the identity of the parties and the authenticity of their statements can be verified beyond doubt.
The allocation of responsibility regarding submissions and draft judgments prepared by generative AI is one of the most pressing dilemmas in legal dogmatics. Responding to Boldizsár Szentgáli-Tóth's inquiry regarding ethical risks and the duty to inform, the participants clarified: the use of AI does not exempt one from professional liability. According to Dóra Pálfi, disclosing AI use to the court would not significantly assist the administration of justice, as judges and opposing parties must subject every submission to the same level of critical scrutiny. The liability of the legal representative or the judge is indivisible; "factual hallucinations" or fictitious statutory references generated by the algorithm constitute professional negligence. The Colombian example mentioned in connection with automated decision-making also confirms that AI is acceptable as an auxiliary tool as long as the essential consideration and final decision are made by a human actor, ensuring the "human-in-the-loop" principle.
The social risks of algorithmic decision-making, particularly the amplification of biases, were also highlighted during the debate. AI tends to automate systemic distortions inherent in the input data, which can lead to indirect forms of discrimination, such as ethnic bias, in the judiciary. Dóra Pálfi warned that technology is far better suited for the retrospective identification of anomalies in judicial decisions than for making future judgments. The full-scale implementation of "machine sentencing" faces not only technical but also constitutional obstacles, as the right to legal remedy and human dignity require that legal disputes be decided by a human being capable of an individual weighing of legal norms and social justice.
Regarding the future of the legal profession, Adrienn Nagy explained that applying public models like ChatGPT to professional work poses serious risks. These systems often yield convincing results only at a stylistic level, while their professional depth is minimal, and verifying the content they generate consumes more resources than their use saves. The path forward lies in specialized legal models built on closed, verified datasets. Digital solutions currently being prepared by the Ministry of Justice follow this path: by building on the complete database of Hungarian legislation and court decisions, they aim to create systems that provide reliable assistance to professional legal practitioners in analysing case law, beyond informing laypeople. The lawyer of the future will thus become a kind of "IT-lawyer," whose core competence will include understanding the logic of algorithms and filtering out technological errors.
Concerning social equality and access to justice, experts nuanced the image of AI's democratizing role. While technology could theoretically assist those who cannot afford a lawyer, Dóra Pálfi pointed out that the "digital divide" may deepen. The usage fees for reliable, up-to-date, and validated AI systems will be high, potentially granting wealthier litigants a significant informational advantage. For lay users, the seemingly coherent but legally flawed answers produced by generative AI can provide a false sense of security, leading to groundless litigation or unfounded resistance to judicial decisions. Adrienn Nagy added that a new professional challenge for lawyers is convincing clients, who have been misinformed by AI, of the correct legal solution.
Finally, in examining the regulatory environment, the roundtable participants addressed the European Union's Artificial Intelligence Act (AI Act), which classified justice systems as "high-risk" systems. Boldizsár Szentgáli-Tóth raised the possibility of a national, court-specific regulation (lex specialis); however, the expert opinion suggests this is currently unnecessary. Dóra Pálfi argued that the consumer protection and transparency-focused approach of the AI Act, complemented by the requirement for human oversight, provides sufficient guarantees. Citing European research from 2025, Adrienn Nagy confirmed that the flexibility of the Hungarian legal system and existing liability rules are capable of managing the technological shift. Real efficiency gains are to be expected not from complex decision-support AI, but from simpler algorithms capable of automating mechanical procedures, such as bad-faith delays in insolvency proceedings.
In summary, AI does not replace legal intuition and ethical responsibility; rather, it demands a new kind of professional vigilance and technological literacy from the legal practitioners of the future.
Building upon the insights shared during this discussion, the dialogue is set to continue with the fifth event in the series, titled "Recent Lessons and Further Challenges in Algorithmic Constitutionalism Research." Scheduled to take place on June 11, 2026, this upcoming forum will shift the focus toward the empirical and theoretical milestones achieved within the broader research project.
_________________________________________________________
This report was prepared with the support of the ELTE CSS Algorithmic Constitutionalism Lendület/Momentum Research Group (LP2024-20/2024), funded by the Hungarian Academy of Sciences.
_________________________________________________________
The views expressed above belong to the speakers and do not necessarily represent the views of the Centre for Social Sciences.