Annex III, Category 8Article 50 — Transparency

EU AI Act for Legal Technology

Most legal tech AI is not high-risk — but the exception is important. AI deployed in courts and tribunals to assist judicial decisions is explicitly listed in Annex III Category 8. And almost every legal chatbot and AI document tool faces Article 50 transparency obligations regardless of risk tier.

High-risk (Annex III Cat. 8)

AI for courts and tribunals

AI that assists judicial authorities in interpreting facts, law, or applying law to facts. Full Article 9–15 obligations apply.

Limited risk (Article 50)

Legal chatbots and AI-generated content

AI that interacts with users or generates legal text. Disclosure and labelling obligations apply — not the full high-risk suite.

Which legal tech products are covered

AI decision support for courts and judges

High-risk · Annex III Cat. 8

Legal chatbots and AI intake assistants

Limited risk · Article 50

AI-generated contracts, briefs, and legal documents

Limited risk · Article 50

AI legal research tools used by lawyers

Minimal risk · No Annex III listing

Contract analysis and due diligence AI (lawyer-facing)

Minimal risk · No Annex III listing

Automated legal advice to the public

Assess carefully · May be Cat. 8 or Art. 50

Key obligations

Transparency for AI interactions and generated content

If you operate a legal chatbot or AI assistant that interacts with users, you must disclose it is an AI before interaction begins. AI-generated legal documents and content must be labelled as machine-generated. This is the primary obligation for most legal tech companies — even those not in the Annex III high-risk category.

Risk management system (for court-facing AI)

AI systems deployed in courts or tribunals (Annex III Cat. 8) must have a documented, continuously updated risk management system covering risks to due process, accuracy of legal interpretation, and the rights of individuals affected by judicial decisions.

Technical documentation

Providers of high-risk legal tech AI must prepare full Annex IV documentation before any EU deployment. For AI used in judicial contexts, this includes the training data sources, known limitations on legal reasoning, jurisdictional scope, and the performance metrics used to validate accuracy of legal interpretation.

Transparency to deployer institutions

Instructions of use for court-facing AI must give deploying institutions clear guidance on the system's intended purpose, the types of legal questions it can and cannot reliably handle, accuracy limitations, and the human oversight mechanisms required. Courts must understand what the AI cannot do.

Human oversight

AI used in judicial decision support must be designed so that the responsible authority can always override, disregard, or reverse any recommendation. The system must not create automation bias — its outputs must be presented as inputs to human judgment, not as decisions. Logging under Article 12 supports auditability of human oversight.

Common compliance gaps in legal tech

Assuming all legal AI is minimal risk

Most legal research and document tools are not high-risk. But AI deployed in or sold to courts and tribunals for decision support clearly is. Legal tech companies sometimes underestimate this distinction, leading to a lack of documentation and no risk management process for their highest-risk deployments.

No Article 50 disclosure in legal chatbots

Many legal intake bots, virtual legal assistants, and document-generation tools are deployed without explicit AI disclosure. The fact that users may suspect they are talking to an AI does not satisfy Article 50 — explicit, proactive disclosure before interaction is required.

AI-generated legal content not labelled

Draft contracts, legal briefs, and compliance reports generated by AI must carry machine-readable markers indicating AI origin. Many legal tech platforms deliver AI-generated documents in formats (Word, PDF) without any labelling — this will be non-compliant from August 2026.

Ignoring jurisdictional specificity in training data

Legal AI trained on case law from one jurisdiction (e.g., England and Wales) may produce unreliable outputs when used in another (e.g., German or Dutch law). Article 10 requires training data to be representative of the intended use domain — a legal research AI deployed across the EU must account for the significant variation in civil law jurisdictions.

Start your legal tech compliance documentation

Nytivo generates Annex IV technical documentation and Article 50 compliance checklists for legal technology companies.

Start free trial
FAQ

EU AI Act for Legal Tech — Frequently Asked Questions

Is our AI legal research tool high-risk under the EU AI Act?

Almost certainly not. AI used to assist lawyers and legal researchers with research, document review, or contract drafting does not fall under Annex III Category 8. That category covers AI systems that assist courts, tribunals, and judges in making judicial decisions — not AI tools used by lawyers to prepare their work. Your legal research tool is most likely minimal risk. However, if the tool generates legal advice directly to members of the public (rather than to lawyers), you face Article 50 transparency obligations as a limited-risk AI system.

Does Article 50 apply to our AI legal assistant chatbot?

Yes. Article 50(1) requires that operators of AI systems intended to interact with natural persons disclose that the person is interacting with an AI — unless this is obvious from context. A chatbot providing legal guidance or intake conversations on your platform must make clear that it is an AI system. The disclosure must be given before any interaction begins and must be explicit, not buried in terms of service. This applies regardless of whether the system is high-risk.

We provide AI to courts and tribunals. What obligations apply?

Annex III Category 8 explicitly lists AI intended to assist judicial authorities in researching and interpreting facts and the law and in applying the law to a concrete set of facts. As a provider of such a system, you have the full suite of high-risk obligations: Article 9 (risk management), Article 10 (data governance), Article 11 (Annex IV technical documentation), Article 12 (logging), Article 13 (transparency to deployers), Article 14 (human oversight), and Article 15 (accuracy and robustness). You must also register the system in the EU AI Office database.

Our AI generates legal documents. Does Article 50 require us to label them?

Article 50(2) requires operators of AI systems that generate synthetic text, images, audio, or video to ensure the outputs are marked as artificially generated. For AI-generated legal documents, contracts, or briefs, you must apply machine-readable labelling to indicate AI origin. The exact technical standards for this labelling are being developed by the European Commission, but the obligation to label applies from August 2026. Purely internal documents used solely by a professional who is aware of the AI involvement may be treated differently, but client-facing outputs require clear disclosure.

When must our legal tech AI comply?

For Annex III Category 8 high-risk systems (AI for courts and tribunals), the deadline is 2 August 2026. Article 50 transparency obligations for chatbots and AI-generated content also apply from 2 August 2026. If your system uses a general-purpose AI model (GPAI) from a provider covered by the GPAI rules, those obligations have been in force since August 2025. Start your Article 11 technical documentation now — retrofitting documentation after a system is live is significantly harder than building it in parallel with development.