AI and the Spanish judiciary

Miguel Torres Mingot
Monero Meyer Abogados, Madrid
mtorres@mmmm.es

Unsurprisingly, as in many other jurisdictions, questions are being raised about whether AI could render judgments and what would happen if judges used AI as part of the decision-making process.

Following Regulation (EU) 2024/1689 of 13 June 2024, which classifies this area as high risk in view of its potential impact on fundamental rights and freedoms, and Royal Decree 729/2023 of 22 August 2023, approving the Statute of the Spanish Agency for the Supervision of Artificial Intelligence (AESIA), the Spanish General Council of the Judiciary (the “Council”) issued Instruction 2/2026, dated 28 January 2026. Its purpose is to provide guidance to judges and establish boundaries for their use of AI, while also ensuring public transparency.

The Instruction applies to all AI systems available for use by judges, including generative AI.

Of the various definitions contained in the Instruction, the following are relevant:

• AI system: any machine-based system that, in accordance with the AI Regulation, is designed to operate with varying levels of autonomy and may adapt after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations or decisions that may influence physical or virtual environments.
• Generative AI tool: any AI system capable of generating content automatically, including text, images, sound, code or other materials, in response to instructions provided by a user.
• Limitations and exceptions: Members of the judiciary may use only AI systems provided by the competent judicial authorities or by the Council, or systems subject to quality control and audit by the General Council of the Judiciary. In this context, the Council is responsible for supervising and monitoring the use of AI systems in all matters concerning the processing of personal data for judicial purposes.

The use of any other AI system outside those channels is permitted only if three requirements are met: (a) the use must be for preparatory or research purposes, such as preparing summaries or translations, or analysing legal, academic or technical sources; (b) the information used for such purposes must come solely from open sources, meaning information accessible to the general public without legal or technical restrictions and which may legitimately be consulted by any person; and (c) under no circumstances may judicial data be entered into these “non-official” AI systems. In practice, these restrictions appear designed to discourage reliance on external tools.

Principles governing judicial use

The Instructions sets out several principles governing judicial use, as follows: 

1. Effective human control: the use of AI must always be subject to genuine, conscious and effective human oversight by judges. Such systems must not operate autonomously in judicial decision-making, the assessment of facts or evidence, or the interpretation and application of the law. 
2. No replacement of judges: AI must not be used to make judicial decisions, assess facts or evidence, or interpret and apply the law. 
3. Judicial accountability: judges remain personally accountable at all times, irrespective of any assistance provided by AI. 
4. Judicial independence: AI-generated outputs must not influence, either directly or indirectly, the court’s freedom of judgement. 
5. Respect for fundamental rights: the fundamental rights protected by the Spanish Constitution and EU law must be respected, in particular due process, equality, non-discrimination and the protection of personal data. 
6. Security and confidentiality: judges must safeguard the confidentiality, integrity and security of information and prevent unauthorised access, misuse or unauthorised transfers. 
7. Prevention of algorithmic bias: judges must take appropriate precautions to identify and prevent algorithmic bias arising from the use of AI systems. 
8. Proportionality and limited use: the use of AI must be proportionate to the intended purpose and limited to circumstances in which it may prove useful and effective in supporting or assisting judicial proceedings. 
9. Training and education: judges are entitled to attend training courses on AI. The Instruction also provides that the Council will organise and provide such courses. 

The Instruction further provides that AI systems may be used only to support judicial work, in particular for the following purposes: (a) searching for information, including identifying applicable legislation, case law and academic legal commentary, and retrieving background documentation; (b) analysing, classifying and structuring information, documents or data contained in court proceedings, for organisational purposes, to facilitate understanding or to support the examination of the case; (c) preparing outlines, summaries or internal working drafts, provided that they are not decision-making in nature and do not replace the personal drafting of judicial decisions; and (d) organisational or ancillary tasks.

The Instruction specifically addresses AI-generated drafts.  Any such draft must be generated in full compliance with the terms of the Instruction using only authorised AI systems, thereby expressly excluding external tools. Each draft must undergo personal, comprehensive and critical review and validation by the judge, who at all times retains sole responsibility for the relevant decision. Drafts may never be treated as automated decisions and may only be generated by judges, who must always be able to amend them.

The replacement, automation or delegation of judicial decision-making, the assessment of facts or evidence, or the interpretation and application of the law are prohibited.  In addition,  the following are prohibited (a) the use of AI-generated outputs that directly or indirectly compromise judicial independence or judges’ freedom of judgement; (b) the inclusion in judicial decisions of content generated by AI systems without critical, comprehensive and personal validation by judges; (c) the use of AI systems to process specially protected personal data or information subject to enhanced confidentiality obligations, except in cases expressly authorised; (d) the use of AI systems for profiling individuals, predicting behaviour, assessing risks or classifying individuals, except in cases expressly authorised; and (f) any other uses contrary to the criteria, guidelines and principles set out in the Instruction.

The Instruction concludes by stating that judges remain fully liable for any breach of its provisions, in accordance with their ordinary duties.

While most lawyers are now familiar with AI, the prospect of putting cases before AI systems still feels unsettling. AI undoubtedly offers advantages in many contexts, but it is neither surprising nor unreasonable to expect human control in the judiciary to remain genuine and effective, particularly in cases involving civil rights.

• AI system: any machine-based system that, in accordance with the AI Regulation, is designed to operate with varying levels of autonomy and may adapt after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations or decisions that may influence physical or virtual environments.
• Generative AI tool: any AI system capable of generating content automatically, including text, images, sound, code or other materials, in response to instructions provided by a user.
• Limitations and exceptions: Members of the judiciary may use only AI systems provided by the competent judicial authorities or by the Council, or systems subject to quality control and audit by the General Council of the Judiciary. In this context, the Council is responsible for supervising and monitoring the use of AI systems in all matters concerning the processing of personal data for judicial purposes.