AI regulation in judiciary highlights limitations

The European regulation on artificial intelligence designates judicial activity as a high-risk use, though its implementation remains incomplete and has areas that are not fully covered, according to a recent report. The regulation sets a baseline, leading to the creation of six key principles intended to guide AI use in the legal sector.

Six principles guide AI use in judiciary

These principles include human primacy, human mastery, human responsibility, data protection, preference for internal systems, and continuous improvement. The first principle emphasizes that no judicial decision can be generated automatically, even with a judge's supervision, and that generative AI tools cannot be imposed on judges during their judicial work.

AI use categorized into three levels

An annex to the charter distinguishes this approach from other organizational principles by categorizing AI use into three levels: allowed, allowed under conditions, or excluded. This classification is applied twice, depending on whether the AI tool is internal (developed or used by the court or ministry) or external.

Differences between internal and external tools

The distinction between internal and external tools depends on factors such as where data are stored, who controls them, and whether they can be used to train future models. For companies, the question is not about specific tools like ChatGPT or Copilot, but about the edition, contract, data location, and whether the publisher uses the company's data to train models.

Obligation to train and monitor AI use

The court has established a specific address for reporting malfunctions and suspicious AI results, a measure that companies would also need to implement. Training on AI tools is mandatory and must be tailored to the functions and tools actually used. This relates to Article 4 of the AI regulation, which requires suppliers and users to support the mastery of AI by personnel without forcing them to guarantee specific levels of proficiency.

Generalization of AI tools requires evaluation

The use of an AI tool is only permitted after it has been confirmed that its benefits outweigh its costs and risks. Many companies, however, activate tools first, often at the publisher’s initiative, and only later consider the benefits. The court's charter includes considerations unique to the judiciary, such as the secrecy of deliberations, the independence of judges, and the risk of repeating past legal solutions.

Charter aligns with broader legal and labor agreements

The development of the charter was submitted to the social committee of administration under the first president. It aligns with an agreement from the Ministry of Justice dated July 10, 2026, on the use and evaluation of digital tools, particularly AI. In companies with at least fifty employees, similar considerations arise, as the Labor Code requires the information and consultation of the social and economic committee regarding new technologies.

Regular review and practical implementation needed

The court commits to reviewing its charter at least every two years, a reasonable timeframe given how quickly AI tools and their conditions can change. Drafting the charter is a legal task, while understanding the tools in use—such as their edition, data location, and functions—is a technical task. These two areas are interconnected, and without the latter, even the best-designed charter may describe a situation that no longer exists. The Court of Cassation suggests that its approach could serve as a model for others.