How Does the AI Act Apply to Justice and Democratic Processes?

Which justice functions are listed?

Annex III point 8(a) lists AI systems intended to be used by a judicial authority, or on its behalf, to assist a judicial authority in researching and interpreting facts and the law and in applying the law to a concrete set of facts. That covers research tools that find and summarise case law, tools that assist in interpreting facts, and tools that assist in applying the law to a case, where the system is used by or on behalf of a judicial authority. A general legal research tool used by a private lawyer, a document management system, or a scheduling tool for a court, without assisting in research, interpretation or application of the law to facts, is assessed differently, and the distinction must be recorded from the actual function and user. The record should describe who uses the system, what it assists with, what output it produces and what decision it influences, so the listing can be tested. A system that assists a judge must be assessed with judicial independence and the right to a fair trial in view, not only with the AI Act classification.

What about alternative dispute resolution?

Annex III point 8(b) lists AI systems intended to be used in alternative dispute resolution, where the system's output is used to decide a dispute or to influence the outcome of a dispute resolution process. That covers mediation, arbitration and other out-of-court processes where an AI system assists in, or determines, the resolution of a dispute between parties. A system that only schedules a mediation, only stores dispute documents, or only provides general information about a process, without influencing the outcome, is assessed differently. The record should describe the dispute process, the system's role, the output and the decision it influences, and should consider the parties' rights, including the right to be heard, the right to challenge the output and the independence and impartiality of the process. A dispute resolution system that produces an outcome without a human decision-maker, or without a route to challenge, raises procedural fairness questions that go beyond classification, and those questions need qualified advice before the system is used.

For the broader regulatory framework, see finance, insurance and critical infrastructure.

Which democratic process functions are listed?

Annex III point 8(c) lists AI systems intended to be used to influence the outcome of an election or referendum or the voting behaviour of natural persons in the exercise of their vote in elections or referenda, other than AI systems whose output natural persons are not directly exposed to, such as tools used to organise, optimise or structure political campaigns from a logistical perspective. That listing covers systems that target, persuade or influence voters, including personalised messaging and content systems designed to affect voting behaviour, while excluding purely logistical campaign tools whose output voters do not see. A system must be assessed on its function and exposure: who sees the output, what the output says, how it is targeted and what effect it is intended to have. The record should describe the audience, the targeting, the content and the intended influence, and should consider freedom of expression, electoral law and transparency duties alongside the AI Act. A campaign tool described as logistical that in fact produces voter-facing persuasive content is within the listing's language, and its classification should be addressed directly.

How do judicial independence and fair trial considerations apply?

Judicial independence and the right to a fair trial are fundamental rights that constrain how AI systems can be used in justice, whatever their classification. A judge must be able to decide a case independently, on the facts and the law, without being directed by a system's output, and a party must be able to understand, challenge and respond to material that influences a decision. An AI system used by a judicial authority should therefore be designed and used so that the judge retains decision-making authority, can access the system's basis and limitations, can disregard its output, and can give reasons that reflect the judge's own assessment. The parties should be told, where the law requires it, that a system was used, what it did and how its output was considered. The deployer should record the oversight, the reasons and the challenge route for each consequential use. A system that effectively decides a case, or whose output cannot be understood or challenged, risks undermining the fairness of the proceedings, even where its classification file is in order.

What transparency and oversight are needed for electoral systems?

Electoral systems that influence voting behaviour must be assessed for transparency, oversight and electoral law compliance together. People exposed to voter-facing content should be able to understand that the content is AI-generated or AI-targeted, who is behind it, and how they were targeted, subject to the transparency duties in Article 50 and the electoral and advertising law of the Member State concerned. The deployer should keep a record of the targeting criteria, the content, the audience and the intended influence, and should monitor for misleading content, deepfakes and manipulation that can engage Article 5 and electoral law. Oversight should include a competent person who can review, pause or withdraw content, and a route for complaints and corrections. A campaign that uses AI to influence voters without transparency, oversight or a correction route risks both AI Act and electoral law findings, and the file should be designed to answer questions from an electoral authority as well as from a market surveillance authority.

Continue with law enforcement AI systems.

What does this guide not decide?

This guide does not decide whether any justice or electoral system is high-risk, does not determine whether any judicial decision or election was lawful or fair, and does not give constitutional, electoral or procedural law advice. Those questions depend on the system's function, the user, the exposure, the decision it influences, and the law of the Member State concerned, applied to the specific facts with qualified advice. The guide is jurisdiction-labelled to the EU and does not describe the electoral or judicial law of any other jurisdiction. Where a justice or democratic process system is to be deployed, confirm the classification, transparency, independence, fairness and electoral position with the competent authorities and legal advisers before use, and keep the assessment with the deployment record.

Frequently asked questions

Continue through this AI Act cluster

• finance, insurance and critical infrastructure
• law enforcement AI systems
• migration, asylum and border control AI
• sector oversight, evidence and impact assessment
• transparency duties for AI
• high-risk AI systems