Artificial Intelligence (AI) is increasingly influencing decision-making, and court judgments are no exception. In fact, even within the judiciary, “the adoption of AI tools, particularly generative AI, is becoming a powerful global phenomenon.”

However, the judicial function is too important and sensitive to permit a laissez-faire approach. This is especially true in the European Union, which is leading the way in setting limits on AI use, particularly to protect fundamental rights, the rule of law, and democratic principles.

Unsurprisingly, Regulation (EU) 2024/1689 (the “AI Act”) classifies – and therefore regulates – as high-risk systems those intended to assist judicial authorities in researching and interpreting facts and law in concrete cases. This is probably the first case of strict regulation of AI use by courts.

Soft Law Regulations and the UNESCO Guidelines

Alongside the first hard law regulations, a wide range of soft law initiatives has emerged. These initiatives take the form of guidelines designed to steer courts in their use of AI systems. Recent examples include the Charter on the Use of Artificial Intelligence Systems and of Personal and Sensitive Data published by the Paris Commercial Activities Court in September 2025; the Recommendations on the Use of Artificial Intelligence in the Administration of Justice issued by the Italian High Council for the Judiciary and the UK updated guidance to assist Judicial Office Holders in relation to the use of AI, in October, and, in particular, the Guidelines for the Use of AI Systems in Courts and Tribunals, published by UNESCO in December.

It is worth noting that UNESCO has been particularly active in this field in recent years. This is evident in the adoption of the Recommendation on the Ethics of Artificial Intelligence and the Global Toolkit on AI and the Rule of Law for the Judiciary, both adopted in 2023.

While these documents vary in their specific contents, they all share the same fundamental objective: preventing the irresponsible use of AI by courts (with specific reference to human rights courts, see Maria Pilar Llorens). Assuming judges will use software tools, the goal is to ensure such use aligns with principles guiding the judicial process.

A brief review of the upheld principles makes it nearly impossible not to endorse them. These principles are grounded not only in law but also in reason, and it is fair to assume that a judge who complies with them uses AI appropriately and presumably exercises the function of jus dicere properly.

A Provocative Question

Against this backdrop, one may ask whether these declarations of principles are truly functional in achieving their objective. This question does not primarily concern their nature as soft law instruments, which, by definition, depend largely on the goodwill of those who apply them. This is a well-known issue affecting all soft law instruments, and much has already been said and written about it, so little can probably be added.

The question is more provocative and may even appear iconoclastic.

One might wonder whether declaring principles and rules to be followed ultimately proves counterproductive. In more brutal terms, to say it with Saint Bernard of Clairvaux, “the road to hell is paved with good intentions.”

To support my viewpoint, I would like to refer to the opening lines of Oliver Wendell Holmes’s The Common Law (1881): “The life of the law has not been logic, it has been experience. The felt necessities of the time, the prevalent moral and political theories, intuitions of public policy, avowed or unconscious, even the prejudices judges share with their fellow-men have a good deal more to do than the syllogism in determining the rules by which men should be governed.”

These words highlight an undeniable yet uncomfortable truth. Judges are tasked with identifying and applying the law, but they are clearly influenced by factors that are not directly related to the law. Their individual readings, experiences, political and religious beliefs inevitably influence how they identify, interpret and apply the law. Although one may seek to reduce judicial discretion as much as possible, a certain degree of freedom remains, within which a judge’s prejudices may have a significant impact.

There is little that can be done about this. Indeed, the only real safeguard against the influence of such prejudices is the requirement for judgements to be based on legal reasoning, which must display sufficient rationality to be, if not shared, then at least accepted by its readers.

Undesirable Effects

These considerations may seem to have nothing to do with the issue of AI.

However, on closer inspection, this is not the case. AI is one of the many tools that judges may use – and increasingly do – to decide disputes. It is a particularly powerful (and potentially dangerous) tool, but it remains just that: a tool.

Throughout their life, judges accumulate decision-making tools. This happens when they read books, study foreign languages (which allow access to foreign legal sources), or use a computer rather than a pen to write judgments (thereby enabling them to write longer decisions). The usage of these tools is not regulated because judges are free to choose which ones to use and how to use them.

The only thing that matters to the legal system is that judges properly exercise their function. In this respect, the rules and soft law instruments concerning the use of AI represent a notable exception, justified by AI’s unique capacity to dehumanize decision-making.

However, it should be noted that if cognitive tools have never been regulated, a regulation may eventually prove to be futile. If a judge abuses the tools, this will be reflected in deficiencies in the legal reasoning of the decision, and the very same deficiencies will expose the judgment to criticism. For example, excessive reliance on the ‘copy-and-paste’ function of a word processor may result in inadequate reasoning: it is not the abuse of the copying tool itself that is censured, but rather its effect on the legal reasoning. The same should apply to the use of AI: irresponsible use may lead to judgments based on incorrect precedents or fabricated norms, etc. Yet consequence would not be a sanction for misusing the tool but rather a critique of the legal reasoning itself.

One could argue that guidelines on the use of AI or similar instruments would make it easier to penalize judges for breaching their professional duties. This argument alone acknowledges the usefulness of such instruments and is undoubtedly important. However, the other side of the coin must also be considered. Given AI’s tremendous power and theoretical capacity to write judgments, providing guidelines on how judges should properly use AI could de facto transform the obligation to make autonomous decisions into a less demanding obligation: namely, to make decisions that comply with the rules governing AI use. In other words, once such rules exist, judges might be content to comply with them to render their decisions unassailable (and it is not even particularly important whether the rules are hard law or soft law). Rather than being required to produce legal reasoning capable of persuasion (regardless of the tools used to produce them), judges might only aim to produce reasoning that, while not violating the AI usage rules, could be considered correct. Consequently, the very foundation of the legitimacy of the judiciary might shift from the ability to develop rational legal reasoning to the ability to correctly use existing tools.

This would be a far less demanding task, and the product would be far less vulnerable to challenge. Contesting a decision would no longer depend on highlighting its disputableness, which is intrinsic to human rationality. Rather, it would also require challenging the AI tools deemed suitable by soft law instruments — and perhaps, in the future, hard law as well.

Conclusion

I strongly doubt that this is the intention of those seeking to regulate (and to limit) the use of AI by courts. Therefore, it would be appropriate to reconsider regulatory and policy approaches to the use of AI, recognizing it for what it truly is: a simple tool – albeit a very powerful one – and not a danger in itself. The real danger lies in judges failing to fulfil their obligations or seeking the most convenient solutions when exercising their role. On the contrary, it must be clear that it is the legal reasoning of the judgment that lies at the heart of the exercise of judicial power – and must continue to do so – as it is the only objective evidence by which the rationality of a judgment can be assessed, regardless of how it came about.

Suggested citation:

Paolo Passaglia, ‘The Use of AI by Courts: Regulation, Guidelines, and … the Road to Hell’ (Comparative Digital Law Blog, 04 February 2026) <https://lawandtech.ie/the-use-of-ai-by-courts-regulation-guidelines-and-the-road-to-hell>.

About the author:

Paolo Passaglia is Full Professor of Comparative Law at the University of Pisa (Italy), where he is the coordinator of the School in ‘Law and Geopolitics’ and the director of the Second Degree Master in ‘Law and Technology of the Digital Ecosystem’. His primary research interests include comparative constitutional law, constitutional adjudication, and ICT law. His most recent books are about the abolition of the death penalty (2021), an introduction to Italian constitutional law in French (2022), and the relationship between law and geopolitics (2026).