EU AI-Act Disclosure Statement

Art. 50 EU Regulation 2024/1689. Last updated: 2026-10-10.

1. Nature of the AI system

Mignuti Chatbot is an AI system that answers questions from the customer's sources as a retrieval-augmented chatbot. It combines:

• Embedding model: OpenAI text-embedding-3-small (US, under EU SCCs)
• Primary LLM: Anthropic Claude Haiku 5.5 (a few bots still on Haiku 4.5) — Anthropic PBC (USA), safeguarded by EU SCCs 2021/914; visitor messages are redacted by rules before transfer (including email addresses, phone numbers, IBANs, card numbers, IP addresses, street addresses and names with a title; names mentioned without a title are not reliably caught)
• Fallback LLM: Microsoft Azure OpenAI GPT-4o-mini (EU region, no training on customer data)
• Per project, the customer can instead enter its own OpenAI-compatible model endpoint. All AI calls of that project then go through it, with no fallback to the models above.

2. Transparency to end-users

Every Mignuti Chatbot widget shows the AI notice as soon as the chat window appears, before the first question: "This chat uses AI. Responses may be inaccurate. Do not share sensitive personal data." — in the widget's language. Visitors can collapse the sentence; the short marker “AI · may make mistakes” then stays visible. For screen readers the notice is linked to the chat window and the input field. Customers cannot switch it off on any plan, including white-label (see terms).

Machine-readable marking (Art. 50(2))

Every answer whose text comes from the AI model is also marked in a machine-readable way: in the answer stream (ai_generated: true), in the widget as the attribute data-ai-generated="true" on the message, and in the database, the API and the data export. Fixed texts, such as the notice that no matching information was found, are not marked as AI-generated. WhatsApp only carries plain text; there the marking lives in the database and the export.

3. Human oversight

The customer (business operator of the widget) is responsible for monitoring AI outputs. Mignuti Chatbot provides: conversation logs, similarity-score flagging, thumbs-down reports, and pending-question inboxes as oversight tooling.

4. Prohibited deployments

Customers agree not to deploy Mignuti Chatbot for:

• Emotion recognition in workplaces / schools (Art. 5(1)(f))
• Biometric categorization based on sensitive attributes (Art. 5(1)(g))
• Predictive policing (Art. 5(1)(d))
• Untargeted scraping of facial images (Art. 5(1)(e))

5. Risk classification

Mignuti Chatbot itself is an AI system with transparency obligations under Art. 50. Customer use-cases may move into high-risk categories (Art. 6) — it is the Customer's obligation to assess and, where applicable, conduct a Fundamental Rights Impact Assessment (FRIA) before deployment.

6. Training data

Mignuti Chatbot does NOT fine-tune or further-train the underlying LLMs on customer content. The LLM providers are contractually barred from training on customer data; provider-side retention follows their respective commercial terms (e.g. the Anthropic Commercial Terms).

7. Contact

AI Compliance Lead: media@mignuti.com