EU AI Act · use case
EU AI Act for Legal Tech and Judicial Assistance AI: High-Risk Requirements
AI that helps a court research or interpret facts and apply the law to a case is high-risk under Annex III(8). What legal-tech vendors and courts must document, and what is carved out.
Classify my system in 3 minutesFree, no account. Pre-filled for this use case.
Risk tier
High-risk
High-risk (Annex III, point 8(a)); a narrow administrative carve-out exists
When it applies
High-risk obligations apply from 2 December 2027.
Regulation (EU) 2024/1689, Annex III(8)(a)-(b); Art. 6(3) narrow procedural task exception
AI intended to be used by or on behalf of a judicial authority to assist in researching and interpreting facts and the law, and in applying the law to a concrete set of facts, is high-risk under the EU AI Act. This is the entry that concerns legal-tech vendors selling case-outcome prediction, legal research assistants, and litigation-analytics tools into courts, tribunals and comparable dispute-resolution bodies, as well as AI intended to be used for influencing the outcome of an election or referendum or the voting behaviour of natural persons.
The Act draws a real line here between tools that genuinely touch how a case is decided and administrative tools that merely make a court run more smoothly. Getting that line right determines whether you are building a high-risk product or a minimal-risk one.
Classification
Why this classification applies
Annex III, point 8(a) covers AI systems intended to be used by or on behalf of a judicial authority, or on their 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, or to be used in a similar way in alternative dispute resolution. A tool that summarises case law, drafts a proposed ruling, or scores the likely outcome of a dispute for a judge or arbitrator to rely on fits this description directly.
The recitals carve out AI tools intended for purely ancillary administrative activities that do not affect the actual administration of justice in individual cases, such as anonymising or pseudonymising judicial decisions or documents, communication between staff, or administrative task allocation. Whether a specific product falls on the assistive-administration side or the researching-and-deciding side is often the single most important classification question a legal-tech vendor will answer.
Annex III, point 8(b) separately covers AI intended to be used for influencing the outcome of an election or referendum, or the voting behaviour of natural persons in the exercise of their vote, with an express exception for AI systems whose output natural persons are not directly exposed to, such as tools used to organise, optimise or structure political campaigns from an administrative or logistical perspective.
Obligations
What you have to do
- Classify the product carefully against the Annex III(8)(a) line: does it help research, interpret or apply the law to a case, or is it purely administrative? Document the analysis and revisit it whenever the product changes.
- Risk management system addressing the specific risk of the tool influencing a legal outcome incorrectly or non-transparently (Art. 9).
- Data governance for the case-law and legal-text corpora used, including jurisdictional coverage, currency of the law, and bias examination (Art. 10).
- Technical documentation and automatic logging sufficient to reconstruct how a suggestion or draft was produced, given the accountability expectations of judicial process (Arts. 11, 12).
- Human oversight that keeps the actual judicial or arbitral decision with the human decision-maker, with the tool’s output clearly labelled as a draft or research aid (Art. 14).
- Conformity assessment, EU declaration of conformity, CE marking, EU database registration (Arts. 43, 47, 48, 49).
- For election or campaign-related tools: confirm and document whether the exception for administrative/logistical tools not exposing output to voters applies, before assuming minimal risk.
Paperwork
Documents to have on file
Get these documents drafted for your system
Run the free assessment, then unlock the Compliance Pack: a PDF report plus editable first drafts of every required document and a 90-day plan. €49 one-time, no subscription.
Common mistakes
Where companies get this wrong
- Calling a case-outcome prediction tool “research software” to stay outside Annex III(8)(a). If judges or arbitrators rely on the score to shape a decision, the administrative label does not change the classification.
- Anonymisation and document-management tools that quietly add a “recommended outcome” feature, drifting from the administrative carve-out into the high-risk entry without a re-classification review.
- Election-related ad-targeting or messaging tools assumed exempt because they are “just marketing software”, when their output is directly exposed to voters and intended to influence voting behaviour.
- No clear internal boundary between the AI’s draft output and the judge’s or arbitrator’s own reasoning in the case file, which undermines both the human-oversight requirement and the transparency of the eventual decision.
FAQ
Frequently asked questions
Is a legal research assistant that just finds relevant case law in scope?
Pure retrieval and summarisation without interpreting facts or suggesting how the law applies to them sits closer to the administrative side, but the line depends on how the output is used. If a judge treats the tool’s summary as effectively answering the legal question, document that use pattern and lean toward high-risk compliance.
Does this apply to private arbitration and mediation, not just courts?
Yes. Annex III(8)(a) extends to AI used in a similar way in alternative dispute resolution, so arbitration and mediation platforms with AI-assisted outcome tools are covered on the same basis as court-facing tools.
What about AI used by law firms to prepare cases, not by the court itself?
Annex III(8)(a) is about AI used by or on behalf of a judicial authority. AI used purely by a law firm to prepare its own case is generally outside this specific entry, though ordinary professional, confidentiality and (if consumer-facing) Article 50 obligations can still apply.
This page is general information about Regulation (EU) 2024/1689, updated 2026-09-25. It is not legal advice; classifications depend on the exact intended purpose of a system. Deadlines reflect the Digital Omnibus adopted in June 2026.