EU AI Act
Our statement under Regulation (EU) 2024/1689 for customers and partners who list Rascasse in their own AI documentation.
This statement is written for customers and partners who need to record Rascasse in their own AI documentation — supplier register, customer questionnaires, AI literacy under Art. 4. It is the web version of the statement we issue as a PDF.
Download for your records:PDF
Most of Rascasse is data processing and statistics. “AI” within the meaning of Art. 3(1) of the AI Act refers to the components below; all of their outputs are aids for professional users.
| Component | Purpose & how it works | Technology / model | Classification |
|---|---|---|---|
| Audience modelscore system | Calculation of affinities, reach, demographics and psychographics from aggregated behavioural signals. Where data is thin, values are estimated statistically (imputation) and labelled as such. | Proprietary statistical and ML methods; monthly refresh | Minimal |
| Rascasse Chat / “Ask AI” | Conversational access to Rascasse data. The language model selects the data queries and answers exclusively from the retrieved data; quoted figures are checked against the data basis. | Anthropic Claude (Sonnet 4.6 / Haiku 4.5) via the Anthropic API | MinimalArt. 50 |
| AI-generated text blocks | Short write-ups of already calculated data (e.g. sponsorship brief, rationales in the partner finder, spell-checking of notes). Marked as AI-generated; the user decides whether to use them. | Anthropic Claude via the Anthropic API | MinimalArt. 50 |
| MCP interface for customer assistants | Customers connect their own AI assistant (e.g. Claude, ChatGPT) to Rascasse data. Rascasse only supplies data through a protected interface; the operator of the assistant is the customer. Access is granted only after a data notice has been confirmed. | Model Context Protocol; model chosen by the customer | Minimal |
| Forecasting modulese.g. TourOps | Statistical estimation of demand and catchment areas for events. The result is planning metrics, not automated decisions. | Proprietary statistical models | Minimal |
| Internal data maintenancenot customer-facing | Classification of new entities (industries, IAB taxonomy) and translation of the interface into 8 languages — each with editorial review by our team. | ML classification, language models, DeepL | Internal |
| Category | Applies? | Reasoning |
|---|---|---|
| Prohibited practicesArt. 5 | No | No subliminal manipulation, no social scoring, no biometric identification or emotion recognition, no exploitation of the vulnerabilities of particular groups. Rascasse works exclusively at audience level. |
| High-risk systemsAnnex III | No | No use in the Annex III areas (biometrics, critical infrastructure, education, employment, essential services and creditworthiness, law enforcement, migration, justice and elections). The field of application is marketing, sponsorship and location planning. Note: under Regulation (EU) 2026/1744 these obligations only apply from 2 December 2027 in any case. |
| High-risk productsAnnex I | No | Rascasse is not a product and not a safety component within the meaning of EU harmonisation legislation (machinery, medical devices, vehicles and others). |
| Transparency obligationsArt. 50 · in force since 2 August 2026 | Yes, met | The chat and text functions are clearly recognisable as AI; AI-generated texts are marked as such. Rascasse produces no deepfakes, no image, audio or video synthesis, and no content intended to inform the public. |
| General-purpose AI modelsArt. 53 et seq. | Not a provider | Rascasse does not develop foundation models of its own. We use models from Anthropic (Claude), which has signed the European Commission’s Code of Practice for GPAI models. |
| Result | Minimal | A system with minimal risk. No conformity assessment, no CE marking, no registration in the EU database, no reporting obligations for deployers. |
“Rascasse (Rascasse GmbH, Berlin) is an audience intelligence system that provides aggregated behavioural data at audience level. It contains AI components with minimal risk within the meaning of Regulation (EU) 2024/1689: statistical audience models and AI assistance functions based on Anthropic Claude, which are subject to the transparency obligations of Art. 50 and are labelled accordingly. The system is neither a prohibited nor a high-risk system; the audience data do not involve processing personal data of identifiable individuals. Decisions based on the data are taken exclusively by our own staff.”
| Obligation under the AI Act | For Rascasse | Note |
|---|---|---|
| AI literacyArt. 4 · since 2 February 2025 | Yes | Staff who use the AI assistance features should understand how they work and where their limits are (section 4). This statement can serve as training material. |
| Deployer obligations for high-risk systemsArt. 26 | No | Not applicable — Rascasse is not a high-risk system. |
| Transparency towards third partiesArt. 50 | No | Arises only if you publish AI-generated texts from Rascasse unchanged — then label them as AI-generated. |
| Registration / reportingArt. 49, 73 | No | No registration in the EU database, no incident reports required. |
| Supplier documentation | Recommended | Keep this statement on file; adopt the suggested wording from section 5. We will inform you actively if the classification changes. |
Ralf Rattay · Chief Product Officer, Rascasse GmbH · info@rascasse.com
This statement describes the situation as of 25 August 2026 to the best of our knowledge; it does not replace legal advice for the recipient’s own organisation. Sources: Regulation (EU) 2024/1689, Regulation (EU) 2026/1744 (Digital Omnibus on AI), European Commission guidelines on the definition of AI systems (February 2025).
Version 1.0 · 25 August 2026 · first edition.