AI Act — limited-risk dossier

Classification

This tool qualifies as a limited-risk AI system under Regulation (EU) 2024/1689 (AI Act). The tool generates draft letters to airlines under Regulation (EC) 261/2004; the user signs and sends them themselves.

Why not high-risk (Annex III)

Annex III §8(a) targets AI systems that assist a judicial authority in researching or interpreting facts and applying the law. This tool does not assist a judicial authority. It is a consumer-facing drafting assistant where:

• The user is the principal and submits the letter themselves;
• Legal references come from a closed, hand-curated set — no free citation generation;
• No automated decision is made that has legal effect for the user without human intervention.

Compliance with limited-risk obligations

• Transparency (Art. 50 AI Act): The tool clearly indicates that the output is AI-generated, which model is used, which citations are included, and from which source.
• No deception: the tool is not positioned as «we do it for you» or as a legal advisory service.
• Hallucination control: citations are not generated by the AI but inserted deterministically from an allow-list. The AI may only rephrase existing citations, not invent them.
• Closed-set discipline: if the AI removes or alters a citation despite instructions, the tool falls back to the unpolished template version.

User-side risk

A Dutch court awarded full procedural costs to the opposing party in 2025 because a litigant submitted an AI-generated document with an ECLI that turned out not to exist (see ECLI:NL:RBOBR:2025:8495, analysis on IT-en-Recht). That's why: closed citation set, an explicit call to the user to verify every reference before sending, and visible flagging when a citation has not yet been manually verified.

Additional information

• Model output is not legal advice within the meaning of the Dutch Bar Act (Advocatenwet).
• No assignment of claims → no Wft/AFM supervision threshold.
• Data controller / contact: see footer (open-source project).