Can Artificial Intelligence Hear a Litigant Within the Meaning of Article 6 ECHR?

Abstract

The increasing use of artificial intelligence (AI) in judicial systems raises a fundamental question under Article 6 of the European Convention on Human Rights (ECHR): can an AI system genuinely fulfil a litigant’s right to be heard? This article develops a functional account of that right as a multifaceted guarantee requiring the opportunity to present and challenge relevant material, effective examination of decisive submissions, clarification where necessary, contextual and normative evaluation, reasoned decision-making, and responsibility for the outcome. Using doctrinal, comparative, and normative methods, the study examines these requirements in relation to risks associated with judicial AI, including opacity, algorithmic bias, hallucinated outputs, limited explainability, and automation bias. The analysis is situated within the EU AI Act, the GDPR, the CEPEJ Ethical Charter, Romanian civil procedure, and comparative administrative-law cases. The article finds that AI may legitimately assist courts through transcription, translation, legal research, document organization, and non-binding summaries or drafts. However, under current legal and technological conditions, AI cannot autonomously perform the hearing in its entirety. Compliance with Article 6 therefore requires transparency, contestability, independent judicial verification, and meaningful human oversight. Final adjudicative authority, reasoned judgment, and responsibility must remain with an independent and accountable human judge.

1. Introduction—Judicial Artificial Intelligence and the Right to Be Heard

The increasing use of artificial intelligence in courts for efficiency (Karypidou et al. 2026; Vidaki and Papakonstantinou 2025) raises a question that cannot be answered solely by examining the technical accuracy of automated systems. Judicial proceedings are not limited to the processing of facts, legal rules, and documents (Surden 2019). They are also communicative and institutional processes through which individuals present their claims, challenge opposing arguments, explain their personal circumstances, and seek a reasoned response from an independent tribunal (Boroi and Stancu 2020, pp. 22–23). Accordingly, the introduction of artificial intelligence into adjudication must be assessed not only on whether a system can generate a legally plausible outcome, but also on whether it can perform the functions associated with the right to be heard.

According to scholars (Russell and Norvig 2022, pp. 778–79), traditional (discriminative) AI acts as an analytical engine that maps category boundaries to classify data and predict outcomes. In contrast, generative AI acts as a creator, mapping underlying data distributions to synthesize entirely new content from prompts. This establishes a core paradigm shift: while discriminative AI analyzes existing realities and risks misjudgment, generative AI synthesizes new realities and risks fabrication or inventing non-existent reality.

Because the failure modes of generative AI may be difficult to detect, judicial systems must exercise particular caution when using such technologies (Martinez 2025). Although human oversight is essential for all AI systems used in legal contexts, a clear boundary must be maintained between AI-assisted adjudication and the replacement of the human judge by AI (Nowotko 2021). This distinction is also reflected in multidisciplinary scholarship on AI and adjudication. Scholars distinguish between instrumental AI systems and decision-making AI systems in the judicial domain, emphasizing the need to preserve effective judicial protection and a human-centered model of e-Justice (Covelo de Abreu 2024). Research on judicial cognition further indicates that algorithmic recommendations may influence judicial decision-making through automation bias and anchoring, creating a risk of overreliance on computational outputs even when a human judge formally remains in control (Gravett 2024). From a computational and cognitive perspective, legal judgment has also been associated with counterfactual reasoning and the evaluation of alternative factual and causal possibilities, suggesting that adjudication cannot be reduced to classification or prediction alone (Pereira et al. 2024).

Existing scholarship has already advanced the proposition that the guarantees of a fair trial may imply a right to adjudication by a human judge (Górski 2023). The jurisprudence of the European Court of Human Rights also places considerable importance on the effective examination of the parties’ submissions (Van de Hurk v. The Netherlands 1994), adversarial proceedings, equality of arms (Ankerl v. Switzerland 1996), oral and public hearings where required (Colozza v. Italy 1985), judicial independence (Guðmundur Andri Ástráðsson v. Iceland 2020; Khrykin v. Russia 2011), and the obligation to provide reasons (Albina v. Romania 2005). This article examines the functional basis for that protection by asking the question in the title.

However, the existing literature has devoted less attention to the functional content of the right to be heard itself: which procedural and adjudicative functions make a hearing meaningful, and at what point AI involvement ceases to assist those functions and begins to displace the human adjudicative activity protected by Article 6. This article addresses that gap by reconstructing those functions from ECtHR jurisprudence and testing them against distinct forms of AI-assisted and autonomous adjudication.

For the purposes of this inquiry, “hearing” is not understood merely in a sensory or literal sense. Nor is a person heard simply because their submissions have been received, transcribed, summarised, or processed by a decision-making system. Within the framework of Article 6 ECHR, hearing a litigant involves a combination of procedural and adjudicative functions (Bîrsan 2010, pp. 496–98, 501–2). These include enabling the individual to present arguments and evidence, ensuring access to and the possibility of challenging relevant material (Mantovanelli v. France 1997), meaningfully examining decisive submissions (Kraska v. Switzerland 1993), responding to ambiguity or new issues, evaluating the case in its individual legal and factual context, and providing reasons that demonstrate that the party’s position was genuinely considered (Hiro Balani v. Spain 1994).

The article argues that current artificial intelligence systems cannot autonomously perform this combination of substantive, communicative, normative, and institutionally accountable functions. AI systems may process extensive case files, identify legal materials, detect patterns, translate documents, summarise submissions, and generate proposed reasoning. These capacities may improve judicial administration and assist judges in managing increasingly complex caseloads (CEPEJ 2018, pp. 64–65). Nevertheless, information processing is not equivalent to judicial hearing. Current systems lack legal understanding, cannot assume responsibility for a decision, and cannot participate independently in the reciprocal, context-sensitive process by which a litigant’s claims are evaluated and answered (Chen 2026).

This conclusion does not require the exclusion of artificial intelligence from judicial activity. The relevant distinction is between AI that assists a judge and AI that replaces, determines, or materially controls the exercise of adjudicative judgment. The compatibility of an assistive system with Article 6 ECHR depends on whether the human judge remains capable of independently evaluating the case, questioning the reliability and relevance of the automated output, disclosing material reliance where necessary, responding to the parties’ arguments, and assuming full responsibility for the final decision and its reasoning.

The article develops this argument through doctrinal, analogical, comparative, and normative legal analysis. It first identifies the principal elements of the right to be heard in the jurisprudence of the European Court of Human Rights. It then maps those procedural requirements onto algorithmic vulnerabilities, including opacity, bias, hallucinated or inaccurate outputs, insufficient explainability, and automation bias. The European regulatory framework is examined to determine the extent to which human oversight, transparency, and accountability are already required. Some examples of automated administrative decision-making illustrate how algorithmic opacity may impair individuals’ ability to understand and challenge decisions. Finally, Romanian civil procedure is used as a national case study to evaluate the significance of adversarial interaction, public hearings, judicial questioning, contextual evaluation, and reasoned human judgment.

On this basis, the article concludes that artificial intelligence may support the administration of justice, but it cannot autonomously hear a litigant within the meaning of Article 6 ECHR. Final adjudicative authority must remain with a human judge who genuinely engages with the parties’ submissions, exercises independent normative judgment, provides intelligible reasons, and bears institutional responsibility for the decision.

2. Methods

This study employs qualitative legal research grounded in doctrinal and jurisprudential analysis. The primary legal materials comprise Article 6 of the European Convention on Human Rights, selected judgments of the European Court of Human Rights, relevant provisions of the EU Artificial Intelligence Act and the General Data Protection Regulation, the CEPEJ European Ethical Charter, and provisions of the Romanian Civil Procedure Code.

The ECtHR judgments were selected according to their relevance to the procedural elements of the right to be heard, particularly effective participation, adversarial proceedings, equality of arms, oral hearings, judicial independence, the examination of decisive submissions, and the duty to provide reasons. Because the Court has not yet adjudicated a case involving an autonomous AI judge, these principles are applied by functional analogy to contemporary uses of AI in judicial decision-making.

The study also examines selected examples of automated administrative decision-making from different jurisdictions. These examples were chosen from different parts of the world because they illustrate recurring concerns regarding opacity, insufficient explanations, limited contestability, automated factual assessment, and unclear attribution of responsibility. They are not treated as direct precedents under Article 6, but as comparative evidence of the procedural risks that may arise when automated systems materially influence legally significant decisions.

Romanian civil procedure is used as a national case study to test how AI-assisted adjudication interacts with rules governing adversarial debate, oral hearings, judicial questioning, evidence assessment, and the delivery of reasoned state judgments. The analysis is normative insofar as it evaluates whether these functions may legitimately be delegated to AI.

The principal methodological limitation is the absence of direct ECtHR case law on autonomous algorithmic adjudication, which requires that the relevant Article 6 guarantees be applied through functional and comparative analysis.

3. The Right to Be Heard Under Art. 6 of the European Convention on Human Rights and Strasbourg Court Jurisprudence

The right to be heard by a human judge, although not expressly enshrined in Article 6 of the European Convention on Human Rights, is indirectly recognized in the case law of the Strasbourg Court (ECtHR), particularly in relation to the principle of adversarial proceedings and the oral character of the proceedings.

In the case of (Van de Hurk v. The Netherlands 1994), the Strasbourg Court held that the party must be able to present its arguments and the evidence on which it relies, and the court must analyze them and respond to them in the reasons for the judgment, even if this cannot be understood as requiring a detailed answer to every argument. In paragraph 59, the court held that “The effect of Article 6 para. 1 (art. 6–1) is, inter alia, to place the “tribunal” under a duty to conduct a proper examination of the submissions, arguments and evidence adduced by the parties, without prejudice to its assessment of whether they are relevant to its decision.” This means the party has the right to present the arguments and the judge has the duty to effectively examine.

The right to a fair hearing also comprises that the court must analyze their arguments and the evidence taken in. In Kraska v. Switzerland (1993), the applicant alleged a violation of the right to a fair trial, claiming the court had not effectively examined his arguments. One of the judges on the panel openly admitted that he had been irritated by the length of the doctor’s memorandum and that, due to a clerical error, he had received the file only the day before and had managed to read only about 30 pages of it.

The fair trial also means the equality of arms and the adversarial procedure. Both parties must be given the opportunity to have knowledge of and comment on the documents, observations filed, and the evidence taken by the other party. The principle requires “a fair balance between the parties”: each party must be given a reasonable opportunity to present its case under conditions that do not place it at a substantial disadvantage vis-à-vis its opponent (Ankerl v. Switzerland 1996; Kress v. France 2001; Nideröst-Huber v. Switzerland 1997).

In other cases, (Ruiz Torija v. Spain 1994), the court explicitly states that a failure to give reasons “can only be determined in the light of the circumstances of the case.” A minor procedural decision requires far less explanation than a final judgment that strips someone of their property or freedom. The countries bound by the Convention have wildly different legal traditions; some cultures favor short, concise judgments, while others write exhaustive legal essays, but in all cases, the core ideas must be presented.

Access to the court means practical and effective access, understanding the procedural steps and, most importantly, the judge’s reasoning for any decision. The refusal to resolve the merits on grounds related to the priority of an international law that does not allow for challenging a decision (Al-Dulimi and Montana Management Inc. v. Switzerland 2016), the lack of a coherent justification (Albina v. Romania 2005), or even the lack of an answer to an important submission (Hiro Balani v. Spain 1994) entails a violation of the right to a fair trial and reveals the fact that, in reality, the party was not heard.

Also, the Court stated that an oral, public hearing, especially when a person is charged with a criminal offense (Colozza v. Italy 1985; Malhous v. The Czech Republic 2001), is a fundamental right under Article 6 para 1, ensuring an applicant can testify, hear evidence, and cross-examine witnesses (Jussila v. Finland 2006). However, this obligation is not absolute. A hearing is unnecessary if a case raises no issues of credibility or contested facts, thereby allowing courts to fairly decide the matter based solely on written submissions (Döry v. Sweden 2002). Conversely, an oral hearing is mandatory if the court must evaluate factors such as personal suffering to determine compensation (Göç v. Turkey 2002). Courts balance the strict necessity of live testimony against procedural efficiency based on the case’s specific facts.

Connected to the fair hearing, in another case (Kammerer v. Austria 2010), a minor offense procedure in which the applicant contested the fine imposed for failing to submit his car to the mandatory periodic technical inspection, he claimed violations of the principles of adversarial proceedings and the right to a fair hearing. During the appeal, the court obtained a technical opinion from an official expert but did not serve it on the applicant or hold a public hearing. The ECtHR nevertheless found no violation of Article 6 §1, emphasizing the limited nature of the dispute, the applicant’s legal representation, and the absence of witness examination. The case illustrates that an oral hearing is not always required where no credibility issues or contested facts arise, whereas its omission may become problematic when the nature or stakes of the proceedings require direct participation.

Especially when a vulnerable party’s situation is at stake, the proceedings must uphold the principle of oral hearing, and the party must have the opportunity to benefit from a professional service. In a case (P., C. and S. v. the United Kingdom 2002), a parent was forced to represent herself in exceptionally complex, highly emotional child protection matters concerning her parenting ability. Given the immense stakes and intricate legal issues, the Strasbourg Court held that effective access to the courts and fairness rendered legal counsel indispensable, thereby constituting a breach of Article 6 of the Convention.

Also, in the case of a procedure filed by a mother to juridically incapacitate her son, the Strasbourg Court held that it breached Article 6 because the proceedings were not truly adversarial, as the person was not heard by a judge (Shtukaturov v. Russia 2008). The courts’ jurisprudence reveals the principle that a person of unsound mind must be allowed to be heard either in person or, where necessary, through some form of representation (Winterwerp v. The Netherlands 1979).

In a special case (Stephan Kucera v. Austria 2025), the ECtHR examined whether conducting a hearing via videolink during the COVID-19 pandemic violated the right to an oral and public trial. The Court held that while physical presence is highly desirable, a virtual hearing is not contrary to the Convention if it pursues a legitimate public interest (e.g., public health), has a clear legal basis, and ensures that the litigant and their counsel can fully participate, view the proceedings, and examine witnesses without technical disruption.

Against this context, the analysis suggests that access to a human judge who analyses the party’s requests, mainly after an oral argument by the party personally or assisted or represented by a legal specialist, has the value of a conventional principle that, although allowing exceptions, reflects one of the guarantees of a fair trial. The human judge can evaluate difficult issues or analyze matters that cannot be solved mathematically, such as legal causation, equity, fairness or proportionality (Chen 2026; Górski 2023; Selçuk et al. 2025).

Such an argument is also closely related to how the parties represent or understand their right to be heard. A scholar (Kinchin 2024) argues, the litigants want to speak before the court because they believe it gives them a meaningful chance to persuade the adjudicator, thereby directly influencing the substantive outcome of the dispute.

Also, the right to be heard is inherently linked to persons’ dignity (Zou and Lefley 2025), as the applicant cannot be treated as an object but as a law subject before the judge (Shtukaturov v. Russia 2008). Being granted a forum to express one’s narrative message, the state recognizes and respects the individual’s human agency and moral personhood, regardless of whether the final ruling is favourable.

4. The Mapping of Article 6 ECHR Jurisprudence to Algorithmic Vulnerabilities

As part of the right to a fair trial, the right to be heard can conflict with the use of AI in adjudication. While the European Court of Human Rights has not yet ruled directly on a case in which an AI system acted as an autonomous judge, its extensive body of jurisprudence under Article 6 provides clear, transversal principles that can be systematically mapped onto algorithmic points of failure (Bårdsen 2025).

The use of AI as an autonomous decision-maker, or even as an unverified judicial assistant, may undermine the core procedural safeguards of the Convention (Dias et al. 2026).

In terms of the classical ECHR Article 6 standards, an independent and impartial Tribunal is, first, a tribunal established by law. In a case where the applicant contested the legality of the tribunal, accusing the serious flaw of the appointment procedure of being due to the interference of the executive, the court stated that the examination under the requirement of a “tribunal established by law” must not lose sight of whether the alleged irregularity in a particular case was of such gravity as to compromise the independence of the tribunal in question (Guðmundur Andri Ástráðsson v. Iceland 2020). The independence refers to the personal and institutional independence necessary for impartial decision-making and is therefore a prerequisite for impartiality. It characterizes: a state of mind, denoting a judge’s impermeability to external pressures, as a matter of moral integrity, and a set of institutional and operational arrangements, involving both a procedure by which judges can be appointed in a manner that ensures their independence and merit-based selection criteria, which must provide safeguards against undue influence and/or unlimited discretion by the other branches of the State (Khrykin v. Russia 2011).

In the case of (Sramek v. Austria 1984), the problematic situation was that the rapporteur and secretary of an administrative court was a civil servant in the regional government, directly subordinated hierarchically to the very official who had represented the interests of the state against the transaction denied for the applicant, resulting in a blatant lack of independence and impartiality. The court stated that there was a breach of Article 6 para 1, and such a situation seriously affected the confidence which the courts must inspire in a democratic society.

Also, in the case of (Sacilor Lormines v. France 2006), one issue of impartiality arose because the panel of judges included a magistrate whose future appointment as Secretary General of the Ministry of Economy and Finance was already under negotiation and analysis during the very deliberations of the case, so the applicant challenged the presence of the Government Commissioner in the deliberations.

Against this background, an unofficial (clandestine) judicial reliance on AI models with opaque algorithms for autonomous adjudication compromises the right to a tribunal established by law, delegating decision-making power to an entity other than the appointed judge. Then, even if an AI system, not specifically designed for the justice system, is officially accepted for adjudication, there is a breach of the independence requirement, equivalent to delegating adjudicative authority to private developers who build the underlying model. Even when the delegation of adjudicatory power is directed to a specifically conceived justice-system AI model, independence and impartiality, in the sense of ECtHR jurisprudence, can be compromised by bias in the algorithm’s decision-making (Javed and Li 2024; Jha 2024).

Also, officially approving AI only as an auxiliary tool creates a risk for judges’ independence. These tools can automate the preparation of certain procedural acts, the generalization and systematization of evidence, the selection of relevant case law, and the preparation of court transcripts (Borgesano et al. 2025). When judges routinely defer to algorithmic outputs, mere assistance degrades into uncritical reliance. Consequently, the constitutional mandate for judges to independently evaluate facts and apply human reasoning is compromised.

Even if the reading and synthesizing capacity of an AI model is incomparable to that of a human, hallucinatory outputs (Mishra and Soy 2026) are not excluded, proof that the system may have processed or represented the material inaccurately, even if it read the whole document provided by the party. AI does not understand the law or human values, but it generates highly convincing natural language based on probabilistic pattern matching (Boretti 2026). This lack of semantic understanding makes it susceptible to “hallucinations,” in which the system invents non-existent legal precedents (Mata v. Avianca, Inc. 2023) or distorts facts with an air of absolute authority (Weiss 2025).

Regarding videoconferencing for the debates, an important distinction must be made: a virtual hearing modifies the spatial medium of a trial but preserves a human judge; automated platforms replace the human judge entirely. While remote proceedings alter the physical geography of the courtroom, they maintain the essential communicative architecture of a fair trial. The litigant still interacts with a conscious human capable of interpreting emotional distress, weighing moral nuances, and engaging in responsive dialogue. Conversely, delegating adjudication to an AI system is not a mere change in format but the elimination of the adjudicator. It reduces the right to be heard from a meaningful human connection into a sterile process of algorithmic data extraction, fundamentally severing the guarantee required by Article 6 ECHR.

In this context, if an AI system replaces the human judge, it cannot genuinely listen to or understand the parties’ emotions, lived experience, or qualitative and narrative arguments, because it processes information through computational mechanisms rather than human cognition. If the AI model cannot prioritize fostering trust and respect between communicators (Polupanova et al. 2026), there is no understanding and no real communication. As the scholars (Stein and Shollo 2025) have shown, although artificial cognition is evolving, it remains tied to information processing, whereas human cognition is not; effective human-AI integration requires rejecting the paradigm of artificial intelligence as a cognitive substitute, and instead promoting “appropriate reliance” (Schemmer et al. 2023), while preserving human judgment, grounded in authentic emotional capacity and lived somatic experience (Kim and Peng 2025), and approaching any transition toward AI adjudication gradually (Linna and Linna 2026). Empirical research similarly shows that people generally prefer limited judicial automation and trust human judges more in determining fair outcomes, although automation is more readily accepted at the information-acquisition stage (Barysė and Sarel 2024). Recent findings, particularly in criminal justice, also indicate continued resistance to AI adjudication, as “the public still demands that justice be rendered by a fellow human being, however flawed” (Shao et al. 2026).

However, the entire paradigm shifts fundamentally when artificial intelligence is demoted from a substitute decision-maker to a strictly supportive assistant. This distinction is paramount because it neutralizes the threat to the right to be heard. When AI is used purely as an assistive mechanism, such as employing speech-to-text models for real-time transcription or algorithms to summarize vast evidentiary records, the AI is stripped of its adjudicative power. The machine handles the computational and administrative burden, but it no longer interprets the human condition. The human judge acts as the essential moral and emotional filter. Because the judge remains actively and physically present, they retain exclusive cognitive authority to contextualize the AI-synthesized text through the lens of human empathy, reading courtroom dynamics the algorithm cannot “see.” The AI organizes the data, but the judge evaluates the parties’ dignity, suffering, and qualitative nuances. By retaining the power to review, correct, or reject AI outputs, the judge preserves both judicial accountability and the human decision-making required by the procedural guarantees discussed above.

There may be a problem when AI replaces the human judge, because the algorithms are opaque (Lawanna 2026); their functioning is not very clear even to providers (Rudin 2019), so litigants cannot challenge evidence or parameters of an AI system if they do not understand them (Metikoš et al. 2026), since the source code, training datasets, and weight distributions are technically unexplainable. In this context, scholars (Mersha et al. 2024) call for explainable AI (XAI) for transparency (Nowotko 2021) and accountability (Carnat 2024; Selçuk et al. 2025) as complex AI models are integrated into people’s lives.

Also, a judge relying entirely on an AI system cannot explain the exact internal logic behind the output (Widłak 2026), leaving the litigant unable to mount a substantive appeal. Judges cannot simply adopt AI-generated text as their own behind closed doors; they must be legally compelled to be completely transparent about the AI’s involvement (Hendrickx 2025), ensuring that the human judge remains accountable for the logic, fairness, and technical limits of the final ruling.

A person’s inability to challenge a decision they cannot understand shows that the right of access to the court is only theoretical and illusory. The “black box” problem of AI (Michaels 2020) is not merely a technical limitation but can be a direct threat to the ECHR’s adversarial principles.

5. The Human in the Loop Regulatory Framework

To prevent the erosion of fundamental due process guarantees, a regulatory framework was developed within the European legal space to ensure human oversight of automated systems.

The most significant legislative instrument is the European Union’s Artificial Intelligence Act (Regulation (EU) 2024/1689). Under Annex III, point 8 (a) of the AI Act, AI systems intended to be used by or on behalf of a judicial authority to assist in researching, interpreting facts and law, and applying the law to a concrete set of facts are classified as “high-risk”.

The AI Act imposes strict, legally binding obligations on both the providers and deployers of high-risk AI systems. Under Article 14, deployers must ensure that these systems are subject to effective human supervision to prevent risks to fundamental rights. Also, according to Article 86 para 1 of the AI Act, if a high-risk AI system (under Annex III, excluding point 2) outputs a decision causing legal, health, safety, or fundamental rights harms, the affected person has a right to clear, meaningful explanations from the deployer regarding the system’s role and the decision’s main elements. This right empowers affected individuals to obtain clear and meaningful explanations of the AI system’s role in the decision-making process and the main elements of the final output.

This statutory framework interacts directly with the General Data Protection Regulation (Regulation No 679 2016). Article 22 para 1 of the GDPR establishes that a data subject has the right not to be subject to a decision based solely on automated processing that produces legal or similarly significant effects. The historical origin of this protection lies in the French Data Protection Act of 1978, which prohibited automated profiling in judicial or administrative decisions (Barros Vale 2022).

At the Council of Europe level, the 2024 Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law (Council of Europe 2024b) introduces an AI-specific framework for the protection of human rights. While its Preamble recognises that activities throughout the lifecycle of AI systems may offer “unprecedented opportunities to protect and promote human rights, democracy and the rule of law”, the Convention subjects such activities to requirements of judicial independence, transparency, accountability, effective remedies, and procedural safeguards. Particularly relevant to adjudication are the requirements that individuals receive sufficient information to challenge decisions made or substantially informed by AI (Article 14) and, where appropriate, be informed when they are interacting with an AI system rather than a human (Article 15). The Explanatory Report (Council of Europe 2024a) further associates rights-affecting AI decision-making with meaningful human oversight, including ex ante or ex post review (Article 15 para 103). The Convention therefore reinforces a model in which AI may support decision-making subject to effective human control, rather than operate as an unreviewed substitute for the responsible decision-maker, although it does not expressly establish a right to adjudication by a human judge. Under Article 30(3), the Convention will enter into force following five ratifications, acceptances, or approvals, including at least three Council of Europe member States; this threshold had not yet been reached as of September 2026.

The relevant Council of Europe and European Union instruments operate within distinct, although interacting, normative frameworks. The ECHR establishes international human-rights obligations binding on its Contracting States, while Article 6 and the Strasbourg Court’s case law define substantive fair-trial guarantees, including the right to be heard, without regulating AI as a technology. By contrast, the AI Act and the GDPR are EU regulations that are binding in their entirety and directly applicable within the Union legal order. Where a situation falls within the scope of EU law, Article 47 of the Charter of Fundamental Rights (The European Parliament 2000) guarantees an effective remedy and a fair hearing, while Article 51 determines the Charter’s field of application. Article 52(3) further structures the relationship between Article 47 of the Charter and Article 6 ECHR: Charter rights corresponding to rights guaranteed by the ECHR must have the same meaning and scope, without preventing EU law from providing more extensive protection. Accordingly, Article 6 ECHR constitutes an important interpretative baseline for the fair-trial guarantees under Article 47, while the latter has a broader field of protection, since it is not limited to disputes concerning “civil rights and obligations” or a “criminal charge”. The distinction also extends to enforcement: national courts may apply EU regulations directly within the Union legal order, whereas compliance with the ECHR is secured through domestic implementation of Convention obligations and the supervisory jurisdiction of the European Court of Human Rights. In the field of judicial AI, the AI Act expressly regulates systems used to assist judicial authorities as high-risk systems and requires effective human oversight; recital 61 further states that final decision-making must remain human-driven. The 2024 Framework Convention operates at a different level by translating general human-rights obligations into AI-specific safeguards. Article 27(2) expressly preserves the application of EU law between EU Parties, showing that the two frameworks are complementary but retain their distinct legal roles and effects.

The CEPEJ European Ethical Charter on the use of Artificial Intelligence in judicial systems and their environment, an important soft-law instrument, sets out five core principles to guide the implementation of AI in judicial systems. While emphasizing fundamental rights, non-discrimination, security, and transparency, it places paramount importance on the “under user control” principle, reinforcing the “human-in-the-loop” doctrine. Subsequent Council of Europe instruments have progressively operationalized that principle. The CEPEJ Assessment Tool (CEPEJ 2023) adopted in 2023 translates the Charter’s ethical principles into concrete checks, safeguards, and institutional measures for AI systems used in justice, while the Information Note on the use of Generative AI (CEPEJ 2024) addresses risks specific to generative AI. The regulatory trajectory therefore reflects a movement from general ethical guidance toward operational safeguards and, through the 2024 Framework Convention, toward treaty-based human-rights obligations once that Convention enters into force.

In this legislative context, the analysis suggests that the use of AI systems for the judge’s specific activity is not excluded, but such an implementation requires concrete legislative measures in each state. Express norms are needed to recognize the possibility for judges to use artificial intelligence, even just as an aid in their work (Hendrickx 2026). The AI Act establishes directly applicable requirements for the development, deployment, and human oversight of relevant AI systems, but it does not determine which adjudicative functions such systems may perform. National procedural law still governs that question and must define judges’ permissible uses of AI and the applicable safeguards.

6. Difficulties in Automated Administrative Decision-Making

As there is still no jurisprudence at the ECHR due to states’ reservations about implementing an AI judge, several cases may be relevant in which AI models are used for decision-making in administrative activity, preceding access to a court.

In the administrative law case of (Haghshenas v. Canada 2023), the applicant challenged an immigration officer’s refusal of a work permit, which was, in part, guided by the “Chinook” software tool. The applicant argued that procedural fairness required the state to disclose how the software functioned and how it influenced the final decision. The court rejected this request, arguing that the final decision pertained to a human officer, highlighting the high barrier litigants face in challenging algorithmic reasoning in judicial review.

In case SyRI (Systeem Risico Indicatie), a socio-technical infrastructure and automated welfare fraud detection system developed in the Netherlands, which allowed various Dutch administrative organs, including the tax authority, social affairs ministry, police, and municipalities, to collaborate and exchange data, the plaintiffs asked the Court at First Instance in The Hague to declare the SyRI legislation unlawful. They argued that the system severely interfered with citizens’ right to privacy under Article 8 of the European Convention on Human Rights and violated data protection standards. The government refused to disclose the exact risk model, the algorithm used, or the Ministry’s analytical methods, claiming that secrecy was necessary to prevent fraudsters from gaming the system. The court invalidated the system and ruled that it failed to strike a “fair balance” between fraud prevention and privacy due to a severe lack of transparency and an overly broad collection of citizen data (Van Bekkum and Borgesius 2021). It warns policymakers that state-run algorithms must comply with data protection principles and maintain strict public transparency.

In Italy, the Ministry of Education introduced an automated system to manage inter-provincial mobility and job assignments for schoolteachers nationwide. Teachers challenged their automated reassignments before the administrative courts, arguing that the algorithm resulted in unfair allocations, split up families, and provided absolutely no explanation for why specific transfers were ordered. The Council of State ruled against the Ministry and held that while the administration can use AI to improve efficiency, it cannot fully substitute human discretion, and an administrative act generated by an algorithm must still fulfil the traditional duty to provide comprehensible, non-arbitrary reasons (CJC Database 2019). The court held that the automated decision-making system infringed upon the plaintiff’s rights by precluding an individualized assessment of her circumstances, thereby generating an inequitable outcome.

In Australia, the Online Compliance Intervention (OCI) system used automated data-matching to compare welfare recipient records with tax data, averaging out income over a year to automatically calculate and issue debt notices to individuals assumed to have been overpaid (Clarke et al. 2024). The system shifted the burden of proof to vulnerable citizens to “disprove” the automated debt. The Federal Court ruled that the automated averaging method was unlawful, forcing the government to scrap the system and refund substantial sums to hundreds of thousands of citizens.

These cases show that the use of AI in administrative decision-making has led to problems with understanding the algorithm or how the system works, generally affecting private individuals with no IT training who, in the absence of a clear explanation, cannot understand or challenge the resulting decision. Moreover, in some cases, private individuals were forced to dismantle the AI system’s results through evidence, which, given the lack of transparency in how it works, was extremely difficult. Therefore, those interested turned to the judiciary, which was called upon to analyse whether there was a motivation for the administrative act and, if so, what the machine’s reasoning was.

When the issue of an administrative act is in question, the assistance provided by AI to a public official may render the act unlawful, in the context in which the statement of reasons for the administrative act, an intrinsic, essential requirement for the validity of the administrative act (Charter of Fundamental Rights of the European Union, Art.41 Para 2 Letter c 2012), is not presented or is not clear. The lack of motivation stems from a lack of understanding of how an AI model works, acts, or resolves issues, given its black-box nature (Liu et al. 2025).

The analogy to judicial proceedings becomes clear: if an algorithmic ‘black box’ renders an administrative act legally void for lack of comprehensible reasoning, its use in adjudication fundamentally nullifies the Article 6 ECHR right to be heard, as true justice requires a transparent dialogue that a machine cannot provide. If the reasoning behind a decision is hidden, the litigant is deprived of the very essence of the right to be heard, the ability to effectively understand the case against them, challenge the underlying evidence, and engage in a truly adversarial debate.

This analogy also supports an a fortiori argument. If administrative decision-making, which is subject to comparatively less stringent procedural guarantees, already requires a meaningful level of human responsibility, intelligible reasoning, and individualized assessment, then judicial decision-making must satisfy these requirements with even greater force. Article 6 ECHR subjects adjudication to heightened guarantees of independence, impartiality, adversarial procedure, and reasoned judgment. Accordingly, if predominantly automated decision-making raises legality concerns in administrative law, substituting the human judicial assessment is even harder to reconcile with the stronger safeguards governing a fair trial.

7. The Romanian Civil Procedure Framework

Romania offers a case study in domestic procedure: an ECHR Contracting State whose Civil Procedure Code, in force since 2013, is modern. Drafted in the digital era, it guarantees a fair trial, yet it still reflects the idea of a human judge, even if only indirectly. The gap with AI is therefore not the result of outdated law: the Code is simply not ready for AI judges.

In Romanian civil procedure, the adversarial principle (Article 14 of the Civil Procedure Code) and the principle of public hearings (Article 17 of the Civil Procedure Code) are not mere technical guidelines; they form the constitutional bedrock of human-centric justice (Tăbârcă 2023). Together, they dictate the general rule that a judicial decision can be forged through open, communicative, and live human interaction, elements that an artificial intelligence judge is structurally blind to.

Article 14 of the Civil Procedure Code requires that all arguments, explanations, and evidence be subjected to mutual, real-time debate. This adversarial exchange relies heavily on the sensory and emotional acuity of a human judge. Legal persuasion is not confined to the textual content of submissions and testimony; it also emerges through paralinguistic and behavioural forms of communication observable during judicial proceedings. During the cross-examination of a witness or the hearing of a party’s personal statement, a human judge may assess variations in tone, response latency, posture, facial expressions, eye contact, and other non-verbal indicators that can provide contextual information regarding the witness’s emotional state, credibility, degree of certainty, or recollective process.

Article 352, paragraph 2 of the Civil Procedure Code explicitly allows the judge and the adversarial parties to extend the scope of the questions beyond the initial written framework (Ciobanu and Nicolae 2013). This fluid, spontaneous expansion means that the emotional state, immediate reactions, and psychological vulnerability of the party being heard become active vectors in uncovering the substantive truth. The human judge can ask follow-up questions, recognize ambiguity, revise an initial understanding, respond to the parties, and, in the end, justify the evaluation, his decision being understandable and susceptible to challenge.

In contrast, an AI system, processing flat transcription data, cannot read these behavioral cues, mistaking semantic consistency for truth or emotional shock for non-cooperation. Speech-to-text applications, tested in some courts, improve celerity and efficiency (Bădoiu et al. 2026), with the clerk being discharged from writing during the public hearing and then creating the document based on the data collected in court. But, as Article 264 of the Civil Procedure Code requires, the understanding and appreciation of one’s declaration stay with the judge, who is compelled to analyse himself and value the evidence taken in only by his own perception and critical thinking (Ciobanu and Nicolae 2013; Decision 948 2023; Decision No 1631 2022).

Simultaneously, Article 17 of the Civil Procedure Code mandates that court hearings be public, generally a vital value for due process (Vidaki and Papakonstantinou 2025). This publicity serves as a democratic mirror, exposing the judicial theatre to public scrutiny. A public hearing is an audible and visible event in which the community witnesses litigants’ vulnerabilities. In complex family disputes, such as child custody or domestic negligence, in proceedings involving vulnerable people, or in labour cases of workplace harassment, the context is deeply emotional. A victim of domestic violence speaking with visible apprehension as she recounts years of abuse, a defendant expressing genuine remorse during sentencing, or an elderly person struggling to maintain composure while describing the loss of a lifelong home conveys dimensions of human experience that cannot be fully captured by transcripts or written submissions alone. These manifestations of vulnerability, fear, regret, or emotional distress provide contextual information that may assist judges in assessing credibility, understanding the personal significance of the events described, and appreciating the broader human consequences of legal disputes. Such observations form part of the immediacy of courtroom proceedings and illustrate the extent to which judicial evaluation often relies on more than the semantic content of spoken words.

As was already shown, Article 6 ECHR does not require an oral hearing in every case, as some disputes may be decided fairly through written proceedings. In this case, the litigant is legally “heard” without appearing physically before a judge. The Romanian Civil Procedure Code provides an exception to the general rule in Article 1030, which states that the small-claims procedure is written and takes place entirely in the council chamber (Boroi and Stancu 2020). The law allows the judge to determine whether the parties’ presence is necessary, but this is the exception; usually, oral argument is not required. In this case, as the debates are only in writing, a question arises: Can an AI system better adjudicate than a human judge if the whole contextual data is presented in the documents filed by the parties? Even if artificial intelligence may process the submissions more consistently than a human judge and detect inconsistencies, in the end, under Article 395, corroborated with Article 425, paragraph 1 of the Romanian Civil Procedure Code, it is the tribunal that must meaningfully present the reasons and the legal reasoning in the decision after assessing the submissions and applying normative legal standards (Decision No 26 2019). It is important that there is no provision about the use of AI in these proceedings, and under the Romanian Civil Procedure Code, the human judge is the only one legally responsible for the decision.

8. Discussion: Can AI Hear a Litigant?

It follows that hearing a litigant requires, at a minimum, an opportunity to present arguments and evidence; the possibility of challenging relevant material; the effective examination of decisive submissions; the possibility of clarification or responsive interaction where necessary; contextual and normative evaluation; a reasoned response; and responsibility for the final decision.

At the outset, because judicial proceedings are governed by law, autonomous AI adjudication or the material involvement of AI in adjudicative functions must be compatible with the applicable domestic procedural framework. The AI Act regulates the conditions under which relevant AI systems may be placed on the market, put into service, and used, whereas national procedural law continues to determine judicial competence, procedural form, the conduct of hearings, fact-finding, and the validity of judicial decisions. The AI Act does not, by itself, confer adjudicative competence on an AI system or alter the procedural rules governing national courts.

In Romania, neither the Civil Procedure Code nor other procedural legislation expressly regulates the use of AI in adjudication, although speech-to-text systems have been implemented in some courts. In the absence of a specific domestic procedural basis, AI cannot presently be assigned an autonomous adjudicative function in Romanian civil proceedings. The human judge remains responsible for the conduct of the proceedings, the assessment adopted by the court, and the content of the final decision. This does not necessarily exclude ancillary or preparatory tools that do not materially determine the assessment of facts or law or the outcome of the case. Their use must nevertheless comply with the applicable procedural, data protection, confidentiality, cybersecurity, and judicial administration requirements.

Where domestic law permits the use of AI, the implications for the parties’ right to be heard must be assessed considering both their opportunity to present arguments and their right to have those arguments genuinely examined by the court. An AI-assisted system may provide valuable support by organizing large volumes of case materials, identifying potentially relevant submissions, detecting inconsistencies, and reducing certain forms of human oversight error. Its usefulness, however, depends on the system’s design, training data, reliability, intended purpose, context of use, and the quality of human supervision.

Under the human-oversight model reflected in the AI Act, the judicial decision-maker cannot be replaced by the AI system. Human oversight requires more than the formal approval of an AI-generated output. The judge must possess the competence, information, authority, and practical capacity necessary to assess the output critically, identify errors and limitations, disregard or override the output where appropriate, and retain effective control over the adjudicative process.

Where an AI output materially contributes to the assessment of facts or law, the judge must independently examine the relevant facts, submissions, and legal rules rather than merely confirm the system’s conclusion. In relation to facts, such a contribution may consist of retrieving relevant information from large volumes of evidence, identifying links between documents, detecting patterns, or suggesting factual inferences. In relation to law, AI may identify potentially applicable provisions, retrieve relevant case law, propose interpretations, or generate draft reasoning. The latter is particularly sensitive because it may shape the judgment’s normative basis. In both contexts, however, the output must be independently checked, verified, and filtered through the judge’s own judgment, since the judge alone bears responsibility for the decision. A Convention-compliant procedural framework should require disclosure sufficient to enable the parties to understand the AI system’s material role and to challenge relevant data, assumptions, criteria, or limitations. The judicial decision must also contain the court’s own reasons for adopting any conclusion materially influenced by the system. This does not mean that the judge must identify or explain every AI output considered and rejected; the obligation should concern uses that materially affect the proceedings, the court’s reasoning, or the outcome.

In judicial review of administrative action, the scope of the court’s reasoning is shaped by the contested act and by the reasons, if any, provided by the administrative authority. The judge must determine whether those reasons substantiate the adopted measure, are coherent and intelligible, and satisfy the applicable requirements of legality and good administration.

The administrative cases examined above suggest that opacity becomes legally significant when it prevents the addressee from understanding the basis of an adverse decision or prevents the court from exercising effective judicial review. The relevant question is therefore not necessarily whether the judge can reconstruct every technical operation performed by the system, but whether the responsible authority can provide an intelligible account of the data, criteria, assumptions, and decisional role that materially produced the contested outcome.

Caution is also required where the reviewing judge uses another AI system to assist in drafting the judicial decision. The judgment must express the court’s own assessment, statement of reasons, and legal arguments. An AI-generated draft cannot substitute for the judge’s independent analysis of the administrative act, the parties’ submissions, and the applicable law.

A court is institutionally different from the administrative authority whose act it reviews. The judge benefits from guarantees of independence and impartiality and is subject to all the requirements set out in Article 6 ECHR. Consequently, the judicial use of AI raises an additional set of concerns. Where a judge relies materially on an algorithmic recommendation without providing the parties with sufficient information about its role and the assumptions underlying its output, the fairness of the proceedings may be impaired.

A judicial decision reviewing an AI-influenced administrative act must itself result from proceedings satisfying Article 6 ECHR. The claimant must therefore have a meaningful opportunity to present the alleged unlawfulness of the administrative decision and to receive a response based on the court’s independent assessment. The mere use of an AI system to organize the claimant’s submissions or to prepare a preliminary draft is not necessarily incompatible with this requirement. Incompatibility arises, however, where the judge delegates the substantive assessment to the system or adopts its proposed outcome and reasoning without independent verification, critical examination, and personal responsibility for the decision.

Where proceedings involve the exercise of judicial discretion, the evaluation of disputed facts or credibility, or a right to an oral hearing, replacing interaction with a human judge by direct interaction with an AI system may not satisfy the standards derived from Article 6 ECHR. It may also raise the question whether Article 6 contains an implicit requirement of human adjudication. Although the ECtHR has not articulated a freestanding right to adjudication by a human judge, such a requirement may arguably be derived from the institutional and participatory guarantees of Article 6, particularly where the proceedings require evaluative judgment, responsive interaction, and personal responsibility for the outcome.

Interaction with an AI system does not, by itself, provide an encounter with a legally authorized decision-maker capable of exercising judicial independence and bearing responsibility for the decision. Moreover, AI systems may produce hallucinated, inaccurate, or insufficiently contextualized outputs. Where such outputs are not identified and corrected through effective human scrutiny, the parties may be deprived of the individualized and reasoned examination of their case required by due process.

Depending on the task and the system involved, AI may process submissions consistently, reduce delays, detect inconsistencies, and facilitate translation and accessibility. These advantages are relevant to the practical effectiveness of the right to be heard. Nevertheless, an AI system cannot itself supply the institutional and relational dimensions of an oral hearing before an independent and accountable judicial officer.

The objection to substitution is therefore not merely that AI lacks human emotion. An artificial system does not possess legal personhood, institutional independence, moral agency, or the capacity to bear judicial responsibility. Where adjudication requires responsive engagement with the parties and a context-sensitive exercise of judgment, these limitations may affect the substance, rather than merely the form, of the right to be heard.

The analysis therefore supports a distinction between AI assistance and AI substitution. Article 6 ECHR does not necessarily preclude the use of AI for transcription, information retrieval, translation, record organization, or drafting assistance, provided that such use does not impair adversarial proceedings, equality of arms, effective participation, judicial independence, or the duty to give reasons. Where an AI system materially contributes to the assessment of facts or law, meaningful human oversight must include independent judicial examination, the capacity to disregard or override the output, disclosure sufficient to permit the parties to contest material assumptions, and a reasoned judicial adoption of any conclusion influenced by the system. By contrast, replacing the judge in proceedings that require evaluative judgment, responsive interaction, or an oral hearing raises more fundamental concerns because responsibility for understanding the parties and determining their rights must remain attributable to an independent and accountable tribunal.

This study has several limitations. First, to date, the European Court of Human Rights has not ruled directly on whether autonomous AI adjudication is compatible with Article 6 ECHR. The analysis therefore derives the relevant standards from established case law concerning judicial independence and impartiality, adversarial proceedings, effective participation, the examination of the parties’ submissions, and the duty to provide reasoned decisions. Applying these standards to autonomous AI adjudication necessarily involves some normative interpretation.

Second, AI capabilities are evolving rapidly. Conclusions based on the present technological limitations of AI systems, including their capacity to perform context-sensitive adjudicative tasks reliably, may therefore require reassessment as the technology develops. Accordingly, the analysis is confined to current legal and technological conditions and does not exclude the possibility that future systems may perform particular judicial tasks differently or more reliably. Nevertheless, technological advances alone would not resolve the distinct legal and institutional questions concerning judicial authority, independence, transparency, accountability, and responsibility for the final decision.

9. Conclusions

This article examined whether artificial intelligence can “hear” a litigant within the meaning of Article 6 of the European Convention on Human Rights. The analysis demonstrates that hearing cannot be reduced to receiving, transcribing, summarizing, or processing submissions. It is a composite procedural and adjudicative function requiring an opportunity to present and challenge relevant material, the effective examination of decisive arguments, clarification where necessary, contextual and normative evaluation, intelligible reasons, and institutional responsibility for the outcome.

Article 6 ECHR does not expressly establish a freestanding right to adjudication by a human judge. Nevertheless, the guarantees developed in the case law of the European Court of Human Rights can be understood as presupposing a tribunal capable of independently engaging with the parties’ submissions, controlling the proceedings, resolving ambiguity, evaluating relevance, and explaining the final determination. On this functional account, and under current legal and technological conditions, AI systems cannot autonomously satisfy all the requirements of a fair hearing.

This conclusion does not support the exclusion of AI from courts. AI may legitimately assist with preparing non-binding summaries or draft materials. Such uses may improve accessibility, consistency, and efficiency. The decisive distinction is therefore not between courts that use AI and courts that do not, but between assistive uses that preserve human adjudicative authority and substitutive or materially controlling uses that displace it.

Where AI materially contributes to the assessment of facts or law, Article 6 compatibility requires transparency sufficient for contestation and genuine human oversight, including independent judicial verification and final decisional responsibility.

The Romanian civil-procedure case study reinforces this conclusion. Whether proceedings are oral or written, the procedural framework assigns to the tribunal the conduct of the proceedings, the evaluation of evidence, the application of legal standards, and responsibility for the judgment. AI may support these functions, but it cannot become an undisclosed or effectively controlling source of the decision or its reasoning. The comparative administrative-law examples similarly demonstrate how opacity, the absence of individualized assessment, and unclear attribution of responsibility may make automated decisions difficult to understand and challenge. Although those cases do not directly determine the requirements applicable to judicial adjudication under Article 6 ECHR, they illustrate risks that become more serious when transferred to the judicial sphere.

The analysis is limited by the absence of European Court of Human Rights case law directly addressing fully autonomous AI adjudication, as well as by the rapid evolution of both technology and regulation. Future research should examine concrete judicial deployments, distinguish among different forms and degrees of automation, and develop auditable standards for validation, disclosure, contestability, and meaningful human oversight.

The central conclusion nevertheless remains: AI may assist a court in hearing a litigant, but under current legal and technological conditions it cannot autonomously perform the hearing in its entirety. Final adjudicative authority and responsibility must remain with an independent and accountable human judge.