EU AI Act for Legal Tech

Annex III(8) covers AI '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, or used in a similar way in alternative dispute resolution.' Sentencing aids, judgment-prediction tools, AI-assisted case research deployed in court contexts are explicitly captured. Pure law-firm-internal research tools are LOW-risk unless they make decisions affecting fundamental rights.

Generally NOT Annex III high-risk for B2B contract analysis. BUT: if used to analyse contracts that affect natural persons (employment, consumer credit, insurance), Article 26(9) FRIA may apply downstream when the analysis feeds decisions. And GDPR + confidentiality + legal-professional privilege still apply throughout. Article 13 transparency to deployers is mandatory regardless of risk tier.

Article 14 requires the high-risk AI to be designed and developed so it can be effectively overseen by natural persons during use. For legal tech that means: (a) the human reviewer must be able to fully understand the system's capabilities + limitations, (b) be aware of automation bias, (c) correctly interpret outputs in context, (d) decide not to use the AI in a particular case, (e) override or reverse outputs, (f) intervene or interrupt operation. UI affordances must be in place — not just policy.

Article 22 prohibits decisions based solely on automated processing that produce legal or similarly significant effects on natural persons, with narrow exceptions (consent, contract necessity, EU/MS law authorisation). For most legal-tech use cases, the practical effect is that the human-in-the-loop must be meaningful, not just a rubber-stamp. Legal-tech vendors should show that their UI actively facilitates substantive human review.

Article 14 oversight templates + decision-trace logging at /transparency (£399/mo). Article 13 transparency-to-deployer disclosure templates. /audit-prep-bundle (£4,950) for full Annex IV + FRIA where needed. Bias detection (£299/mo) for any tool ranking judgments or precedents.