ABSTRACT

1 INTRODUCTION

AI is transforming justice worldwide. In 2019, a Chinese litigant attended a trial entirely online—with evidence, witnesses, a judge, and a ruling all conducted through a virtual platform guided by AI. In that proceeding, the human judge relied not only on digital records but also on AI-generated case summaries and precedent recommendations.¹ The verdict was delivered within minutes.

Half a world away, a defendant in Wisconsin, USA, filed a motion for post-conviction relief after learning that the original decision had been influenced by a proprietary algorithm (COMPAS) that neither they nor their counsel could question or understand.² When challenged, the software’s creators refused to disclose how the system made its decisions, claiming trade secrecy.

Similarly, in Canada, the federal Post-Conviction Risk Assessment tool—used to inform probation decisions—was shown to produce racially disparate results, assigning

disproportionately higher risk scores to Black offenders compared to white offenders.³ These cases highlighted the changing role of AI in the justice system: one driven by digital convenience, the other shadowed by intentional and inherent opacities.⁴

In Europe, Estonia briefly became the subject of international media speculation in 2019 when reports claimed it was developing a “robot judge” to decide small claims disputes. The Estonian government later clarified that no such fully autonomous judicial project was underway and that its efforts were limited to exploring digital tools to improve efficiency in the justice system.⁵ This episode illustrates how quickly narratives concerning AI in adjudication can drift from reality, feeding public anxieties about automation in the judiciary.

Yet all four cases reflect a deeper transformation quietly underway in courtrooms across the world. AI is no longer a speculative concept in law; it has become a practical tool used to evaluate evidence, predict outcomes, streamline caseloads, and increasingly, to assist in adjudication. From automated case sorting in the Netherlands to predictive analytics in UK asylum rulings, and from Tax Foresight in Canada to algorithmic mediation in Singapore’s small claims tribunals, AI is beginning to shape who receives justice, when, and on what terms.

But with this transformation comes a set of paradoxes. While AI is expected to resolve judicial backlogs and democratise legal access, it also presents significant challenges to fundamental rights: human dignity, fairness, and open justice.

2 METHODOLOGY

This research adopted a doctrinal legal methodology to examine the integration of artificial intelligence in European judicial systems, specifically in courts and dispute resolution processes. It addresses growing scholarly and regulatory concerns about whether AI can enhance the delivery of justice without compromising foundational legal values, including judicial independence, transparency, human dignity, and the right to a fair trial. The study is particularly timely, given the recent proliferation of AI tools designed to assist or, in some cases, replace judicial decision-making in Europe. The following core objectives guide the research:

• To examine how far the use of AI in court decision-making aligns with legal principles supporting justice in Europe.
• To assess the extent to which artificial intelligence can replace human judges in judicial systems.
• To evaluate whether the integration of AI into judicial processes can enhance the quality and effectiveness of justice delivery without undermining core legal values.

To achieve these objectives, the study conducted a thorough analysis of primary sources, including the EU Artificial Intelligence Act, the European Convention on Human Rights (ECHR), and case law from US and Chinese jurisdictions, with particular attention to Articles 6, 13, 14, and 45 of the ECHR. It also examined relevant domestic and international soft law instruments, such as the CEPEJ Ethical Charter, the OECD AI Principles, and UNESCO’s Recommendation on the Ethics of AI, selected for their influence on the EU AI Act and their guidance on court-level practices.

Secondary sources—including journal articles, books, dissertations, conference papers, newspaper articles, reports, and blogs—were critically analysed to provide context, assess current scholarly debates, and identify normative gaps. These sources were collected from legal databases and academic repositories and evaluated using the CRAAP test.

The methodology incorporated multiple complementary approaches: A comparative legal analysis to examine regulatory and judicial frameworks across Europe, the US, and China, highlighting similarities, divergences, and lessons learned. A systematic legal method, analysing the EU AI Act alongside ECHR provisions and relevant case law to provide a structured assessment of legal norms, principles, and obligations. A critical legal method, assessing the ethical, procedural, and human rights implications of AI in judicial decisionmaking, identifying risks, and proposing normative and regulatory recommendations. This layered approach, combining doctrinal, comparative, systematic, and critical methods, supported an interpretive, evaluative, and normative analysis: interpreting legal norms and jurisprudence, comparing institutional frameworks across jurisdictions, and evaluating the adequacy of current legal regimes.

While no empirical data collection was conducted, illustrative global case references were used to contrast regulatory approaches and highlight practical concerns. This methodology enables the research to provide a well-reasoned, independent legal perspective on the role of AI in judicial systems.

3 RESULTS AND DISCUSSION

AI systems are increasingly permeating the judicial sector worldwide—not only in terms of digital case management but also in supporting core legal tasks such as legal reasoning, data retrieval, and procedural structuring. While the use of AI as a substitute for judicial

decision-making—popularly referred to as “robot judges”—remains highly contested, its application in a supportive or assistive role has gained greater legitimacy in contemporary legal research. As recognised and encouraged by the CEPEJ, such supportive uses of AI can contribute to procedural efficiency, broaden access to justice, and strengthen institutional accountability.

3.1. AI’s Transformative Potential in the Judicial Sector

As AI systems evolve, their potential to streamline courtroom procedures becomes more apparent. In its assistive role, AI can contribute to a wide range of courtroom and pre-trial tasks,⁶ including drafting judgments, intelligent case routing, transcription and translation of legal documents, anonymisation of judgments, and integration with national e-Government systems for the verification of documentary evidence.

Particularly notable is AI’s potential in managing repetitive, low-value claims through online dispute resolution (ODR) platforms⁷—a development that can reduce case backlog without excluding legal recourse.⁸ Moreover, AI can assist judges in technical domains, such as the calculation of financial penalties, the evaluation and distribution of property in family disputes, and the identification of applicable statutes or precedents through efficient legal research algorithms. For example, natural language processing (NLP) models can be used to analyse the factual matrix of cases, and those can suggest relevant legal provisions with remarkable speed and precision.⁹

While these tools do not exercise discretion, they enhance the accuracy and consistency of judicial outputs, contributing to the predictability of legal decisions—an important component of legal certainty and the rule of law. The breadth of AI applications in LegalTech is supported by recent data from the CEPEJ-AIAB, which reported that 125 AI and cyberjustice tools¹⁰ are currently in use or undergoing testing across various European jurisdictions.

These tools, in turn, present distinct benefits and regulatory challenges. For instance, tools supporting document discovery and legal research undoubtedly enhance efficiency but can also amplify existing biases if not regularly audited. Similarly, ODR platforms improve accessibility but raise fairness concerns when human oversight is minimal. Moreover, AIbased predictive tools can support in organising and prioritising cases; however, their heavy reliance on previous datasets risks uniform adjudication, which can undermine individualised justice.¹¹ In addition, decision-support systems that propose sentencing or summarise case facts should augment, not substitute, judicial reasoning; otherwise, they risk diminishing the role of human conscience in adjudication. At the same time, anonymisation tools help safeguard data privacy but can inadvertently hinder transparency in legal scholarship and appellate review. Equally important, translation and transcription tools must be highly accurate, as even slight misinterpretations can lead to unfair outcomes.

Nowadays, NLP models are also reshaping the broader landscape of legal practice. Various law firms already employ either customised or general generative artificial intelligence (GenAI) tools.¹² Between 2023 and 2024, these models became increasingly recognised for their ability to generate human-like text, making them valuable assets for tasks such as legal document drafting, contract analysis, and the provision of preliminary legal advice.¹³ Additionally, the integration of GenAI has expanded the range of AI-supported functions in legal workflows, including summarising complex legal content, producing tailored documents, and facilitating interaction through question-answer systems.

Open-source generative AI platforms, such as those evaluated by French institutions,¹⁴ are also contributing to increased transparency and usability by offering clearer insights into the architecture and governance of generative models. In the private legal sector, GenAI tools are already being explored for their potential to improve productivity and reduce routine workload. Although deployment within judicial institutions remains cautious due to higher ethical and procedural standards, the commercial legal sector has seen a growing interest in adopting these systems for case preparation, legal research, and client communication. As user involvement in system design and testing increases, generative AI is likely to become a practical instrument for enhancing the speed, efficiency, and accessibility of justice delivery.

Although these tools are efficient and automate court procedures, this automation does not come at the cost of due process.

3.2. Ethical Guidelines for AI in the Justice System

Moving forward, it is important to consider the ethical standards guiding AI’s integration into the judicial process. The European Ethical Charter on the use of AI in judicial systems, adopted in 2018, articulates five ethical principles to guide the responsible integration of AI, often referred to as LegalTech¹⁵ within judicial institutions.¹⁶ These principles, though not binding, have emerged as soft law instruments that have influenced both national legislation and court-level policy reforms across Europe.

As the integration of AI tools in the judicial process becomes more widespread, attention must be paid to their ethical governance. Although AI tools promise significant procedural benefits, their deployment must remain aligned with constitutional principles. As judicial reliance on algorithmic systems increases, legal systems must simultaneously reinforce mechanisms for auditability, human review, and accountability.

While current AI systems cannot substitute for judicial conscience or legal reasoning, they can serve as valuable co-pilots by streamlining processes, improving institutional performance, and broadening access to legal remedies when governed ethically.¹⁷

AI technologies are being deployed in adjudication in various ways: as assistive tools supporting judges, procedural aids managing administrative or evidentiary processes, and, more controversially, in fully autonomous roles that aim to replace human judicial discretion altogether. While each of these applications raises legal and ethical questions, the risks intensify as AI moves from assistive to fully autonomous functions.¹⁸

3.3. Assistive AI in Judicial Decision-Making

Despite their potential, assistive AI systems show varied levels of prediction accuracy. While deep learning and large language model–based tools can achieve around 85–90% accuracy on benchmark datasets for tasks such as charge prediction and legal article recommendation, their performance drops to 70–80% in more complex tasks like sentencing and multi-label judgment prediction.¹⁹

At the same time, this integration of AI in court processes raises substantial human rights risks, particularly when individuals have no viable alternatives.²⁰ Although the idea of AI assisting judges in sentencing, bail, or probation decisions offers greater efficiency in justice delivery, it can pose significant threats to the right to a fair trial,²¹ due process, the right to an effective remedy, transparency, and protection against discrimination, while undermining the obligation of courts to provide reasoned judgments.²²

AI systems used as assistive risk assessment tools are only as fair as the data on which they are trained. When fed historical data embedded with structural biases, these systems are prone to replicating and even amplifying those injustices. The phenomenon of the “feedback loop” further exacerbates the discriminatory potential of AI-driven judicial systems. When sentencing algorithms are trained on biased historical data, they can disproportionately impose harsher penalties on certain demographic groups. As these biased precedents are continuously fed back into the system for retraining,²³ they reinforce and perpetuate existing discriminatory patterns.²⁴

Unlike algorithms, human judges can identify and rectify errors, prevent the same mistakes from being repeated in future adjudications, and be held accountable for discriminatory practices.²⁵ This danger is well illustrated in the case of State of Kansas v. John Keith Walls (2017), where the appellate court held that the defendant must be granted full access to the LSI-R (Level of Service Inventory-Revised) assessment, which the lower court had relied upon to determine the conditions of his probation. Denying the defendant access to this assessment prevented him from disputing the accuracy of information that played a critical role in the adjudication,²⁶ thereby violating his constitutional right to procedural due process.

By citing the case Kansas v. Easterling, the court concluded that the district court’s refusal to provide the complete LSI-R report violated the defendant’s constitutional right to

procedural due process during the sentencing stage of the criminal proceedings.²⁷ Such risks show how even assistive AI can compromise fairness when defendants cannot understand or contest algorithmic reasoning.

3.4. Procedural AI in Court Administration

AI is also increasingly used in procedural roles, such as managing evidence, facilitating e-discovery, scheduling hearings, and providing translation or interpretation. While these applications appear less controversial, they directly affect the right to effective participation in proceedings. For instance, effective participation, as emphasised consistently by the ECtHR, involves more than physical presence; it requires that the accused fully comprehend, respond to, and contest the proceedings. If an AI translation system fails to provide accurate interpretation, or if automated scheduling denies adequate preparation time, the fairness of the trial is compromised. Even seemingly minor procedural issues, such as inadequate courtroom acoustics,²⁸ can impede participation. Procedural AI must therefore be critically evaluated to ensure that its efficiency gains do not come at the expense of defendants' rights.

3.5. Fully Autonomous AI (“Robot Judges”)

The most serious challenges arise when AI is deployed as a fully autonomous adjudicator, replacing human judicial discretion. While human judges are entrusted with evaluating factors such as an offender’s intent, remorse, and socio-personal circumstances,²⁹ AI systems are inherently devoid of emotional intelligence and contextual sensitivity.³⁰ By relying exclusively on quantitative metrics and pre-programmed logic,³¹ such systems risk reducing the rich complexity of human conduct into inflexible algorithmic determinations.³² For instance, an AI judge could impose an identical sentence on a coerced, first-time juvenile offender and a habitual adult recidivist, merely because the statutory elements of their offences align.³³ This mechanistic approach, lacking the capacity to discern the moral gravity or mitigating factors of each case, not only generates

disproportionate and potentially unjust outcomes but also undermines judicial discretion, proportionality, and substantive equality before the law.

The principle of judicial independence is also directly threatened by fully autonomous adjudication. Robot judges can inadvertently shift discretion from judges to engineers and programmers who design these systems, thereby hollowing out the independence of the judiciary. In such scenarios, the tribunal is no longer “independent” in substance, even if formally constituted under law. The independence and neutrality of verdicts rendered by robot judges can be jeopardised by even minor changes in the datasets on which they rely; a slight modification by a programmer could entirely alter the outcome of a case. Such vulnerability undermines the very concept of an impartial and independent tribunal.³⁴

A particularly pressing concern is the absence of explainability in algorithmic decisionmaking. AI tools often function as opaque “black boxes”,³⁵ generating outcomes that are neither transparent nor comprehensible to defendants or their legal counsel.³⁶ This deficiency contravenes the core principles of adversarial proceedings and the equality of arms by placing the defence at a marked disadvantage.³⁷ Judgments rendered by robot judges risk violating Article 45 of the ECHR, which mandates that “reasons shall be given for judgments as well as for decisions declaring applications admissible or inadmissible.” If a convicted person cannot comprehend the ratio decidendi, the legal reasoning underpinning the judgment, they are effectively deprived of the opportunity to exercise their right of appeal.³⁸ This directly threatens Article 2 of Protocol No. 7 of the ECHR, which guarantees that “everyone convicted of a criminal offence by a tribunal shall have the right to have his conviction or sentence reviewed by a higher tribunal.” Without a clear understanding of how and why a decision was reached, the exercise of this right becomes illusory, eroding one of the fundamental safeguards of the criminal justice system.

Courts occupy a uniquely sensitive role in upholding the rule of law and protecting fundamental rights. Unlike AI applications in commercial or administrative domains, judicial AI must operate in strict conformity with the constitutional guarantees enshrined

in regional legal frameworks.³⁹ The deployment of AI in this context, where decisions can directly impact fundamental rights and judicial independence, demands legal instruments that extend beyond general-purpose technological governance.⁴⁰ Notably, empirical data from the AI on Trial project indicates that 46 disputes involving the use of AI technologies across various sectors have already reached European courts.⁴¹ In some instances, overly restrictive or innovation-averse judgments have risked stifling the development of beneficial AI tools,⁴² including those designed for legal research, document management, and procedural efficiency. These developments highlight the urgent need for a coherent, robust legal framework to govern the use of AI across all sectors, and particularly within the judiciary.

3.6. EU AI Act (2024) and Soft Laws

The recent EU AI Act, the first comprehensive regulation on AI, adopts a structured riskbased regulatory model, under which AI systems intended for use in the administration of justice are designated as “high-risk”.⁴³ This classification is not arbitrary; it reflects the recognition within the Act that such systems can produce flawed predictions,⁴⁴ introduce biases, and even generate misleading⁴⁵ or hallucinated content⁴⁶ when applied to legal reasoning or case-specific recommendations.

These concerns are not merely technical; they implicate foundational legal principles such as legal certainty, procedural fairness, and the right to a fair trial (as mentioned above). By explicitly acknowledging the serious impact judicial AI can have on democratic institutions and individual rights, the regulation signals that the administration of justice requires heightened vigilance where automation is introduced.⁴⁷

In doing so, the Act draws an important conceptual line between AI tools that influence judicial decision-making and those that serve administrative or ancillary purposes. AI systems that assist in the interpretation of facts, the application of law, or the adjudication of disputes—whether in formal courts or ADR mechanisms—are subjected to greater regulatory oversight, particularly when their outputs bear legal consequences. This distinction reflects a principled commitment to preserving human discretion within the judicial process.⁴⁸

The regulation appears to endorse the use of AI as a supportive instrument, but not as a surrogate for judicial conscience or interpretive reasoning.⁴⁹ Notably, the Act exempts systems designed solely for administrative tasks, such as anonymisation or internal document handling, from the high-risk category, demonstrating regulatory proportionality and avoiding undue constraint on harmless innovation.⁵⁰

Moreover, the AI Act explicitly embodies the normative ethos of earlier soft law instruments such as the EDDRP,⁵¹ the CEPEJ-AIAB guidelines, and the recommendations of the EU AI-HLEG,⁵² despite their non-binding nature. Although these bodies operate independently, they converge around key ethical principles applicable to high-risk AI systems, particularly those deployed in the justice sector.

The principle of human agency and oversight, consistently emphasised in soft law frameworks, is mirrored in the AI Act’s requirement that final legal determinations remain within the remit of human judges. By mandating human oversight, the AI Act acknowledges that judicial independence is not merely formal; it is functional and interpretive, demanding

sustained cognitive engagement from the judge.⁵³ Overreliance on AI-generated recommendations risks diminishing the judge’s active role in legal interpretation and could, over time, degrade the very faculties upon which just adjudication depends.⁵⁴

Thus, the hybrid adjudication model endorsed by the AI Act—a collaborative interface between human judgment and algorithmic assistance—aims not only to safeguard legality but also to preserve public confidence in the judiciary.⁵⁵ A first-ever international legally binding treaty on AI under the Council of Europe has already been signed by 46 countries.⁵⁶

3.7. International Legal Instruments

Global soft law instruments play a crucial role in the governance of AI, complementing hard-law frameworks. Though non-binding, these principles have significantly informed legislative agendas and regulatory discourse, particularly within jurisdictions committed to fundamental rights, democratic governance, and responsible innovation.

A widely accepted framework outlines five key principles that underpin the concept of trustworthy AI.⁵⁷ From a judicial perspective, these principles collectively form a normative foundation that complements European regulatory instruments. The emphasis on transparency and explainability addresses the legal need to understand how AI systems reach decisions. By promoting human oversight, the framework safeguards judicial independence amid increasing automation. The principle of robustness and accountability principles further advocate traceability and risk assessments to uphold fairness. Together, the OECD principles serve as a coherent ethical blueprint, reinforcing the broader international consensus around the responsible use of AI.

Another international framework presents ten core principles that demonstrate an effort to embed human rights, sustainability, and democratic accountability into the full lifecycle of AI systems.⁵⁸ Even though these recommendations are non-binding, they establish universally accepted criteria for the use of AI in an ethical manner, especially in the judicial sector, where the use of technology must respect fundamental human rights. Of particular importance is the principle of proportionality and the mandate to “do no harm,” which

echoes the European risk-based regulatory approach by requiring that AI uses remain strictly necessary to achieve legitimate objectives.

Similarly, the emphasis on human oversight and determination reinforces the imperative that automated systems must never substitute for human judicial discretion but rather remain subordinate to it. In judicial contexts, this ensures that the legitimacy of decisions remains rooted in legal reasoning rather than algorithmic inference. Transparency, explainability, and accountability also form core elements of the UNESCO framework. These principles are especially salient for courts, as they ensure that litigants can understand, interrogate, and, where necessary, challenge AI-influenced outcomes, an essential safeguard for preserving adversarial proceedings and the right to an effective remedy and trial.

Moreover, the Recommendation's insistence on fairness and non-discrimination reflects the growing awareness that AI systems can entrench or amplify structural biases if left unchecked, particularly in areas such as criminal sentencing, bail, and asylum determinations. The inclusion of sustainability as an evaluative criterion further expands the normative scope of AI ethics, underscoring the responsibility of AI actors to consider not only immediate harms but long-term legal and societal consequences. 4 CONCLUSIONS

4 CONCLUSIONS

Undoubtedly, AI has emerged as a powerful tool for enhancing judicial efficiency, and many jurisdictions have begun integrating AI into courts, primarily in supportive and assistive roles. While these applications show significant potential, they also present considerable legal and procedural challenges. Even when AI is used in an assistive capacity—particularly in tasks such as natural language processing (NLP) and generative AI—issues remain, including the production of inaccurate or legally misleading information. Although human judges can correct such errors, their recurring nature highlights the current fragility and limitations of AI systems.

The risks posed by AI differ depending on its role. In an assistive capacity, AI can support judicial decision-making, provided human oversight, transparency, and interpretability are maintained. In contrast, the deployment of AI in a fully autonomous capacity raises substantial legal concerns. Autonomous AI decision-making could compromise core principles of European law, including the right to a fair trial, judicial independence, the delivery of reasoned judgments, the right to appeal, and principles such as transparency, non-discrimination, and equality before the law.

These risks render the full replacement of human judges by AI both problematic and potentially inconsistent with the rule of law—although future improvements in AI design, explainability, and regulatory oversight may mitigate some of these concerns. Current frameworks, including the EU AI Act, still lack clear guidance for judicial actors on how to

integrate AI outputs into decision-making responsibly. This ambiguity creates gaps in algorithmic accountability, explainability, and procedural fairness.

To address these gaps, judicial councils and legal institutions should establish comprehensive guidelines defining when, and to what extent, AI-generated information may be relied upon in courts. Until AI systems can consistently meet the standards of legal reasoning, accountability, and human dignity, their role in adjudication should remain strictly assistive. Fully autonomous AI decision-making, if pursued at all, should be approached cautiously, with risks carefully managed.

FOOTNOTES

REFERENCES

AUTHORS INFORMATION

RIGHTS AND PERMISSIONS

EDITORS

ABOUT THIS ARTICLE

DETAILS FOR PUBLICATION

AI DISCLOSURE STATEMENT

АНОТАЦІЯ