Legal · AI Act
Transparency on artificial intelligence
Regulation (EU) 2024/1689 (AI Act) · art. 50 in force from 2 August 2026 · last updated: 23 September 2026
1. LUMEN uses artificial intelligence — declared in the clear
LUMEN is a financial-analysis system that uses generative AI models to produce reports, models, decks and dashboards. Every output is generated with the assistance of an AI model and states this nature both in the interface (a “Generated by AI” label on the report) and at the foot of the exported documents (PDF, Excel, PowerPoint, Word). The information is shown to the user on first use, as required by art. 50 of the AI Act.
2. Which model and who is responsible
- Base model (GPAI): Claude by Anthropic, invoked via API with the user's key (BYOK). The obligations proper to the general-purpose model provider remain with Anthropic.
- LUMEN's role: deployer (professional user) of the model within its own system. The service controller is the entity that operates LUMEN, indicated in the contact details in the platform and on the invoice.
- Machine-readable marking: the textual outputs rendered in the platform are marked with a provenance marker (a dedicated data-ai-generated attribute and meta tags) that is automatically readable, in line with the detectability obligation of art. 50.
3. Risk classification: LUMEN is NOT “high-risk”
Annex III, point 5(b), of the AI Act qualifies as high-risk the systems used to assess creditworthiness or establish the credit score of natural persons. LUMEN analyzes exclusively legal entities (listed and private companies) on the basis of their accounts and market data: it does not profile natural persons, does not assign credit scores to individuals, does not make automated decisions about credit, employment, essential services or people's rights. Consequently LUMEN does not fall among the high-risk systemsof Annex III and is not subject to the Chapter III obligations set for them. The transparency obligations of art. 50 remain applicable, which LUMEN complies with.
4. Human oversight and intended use
- Mandatory human-in-the-loop. LUMEN's outputs are analysis-support material and require review by a professional before any decision-making use. They are not, and do not replace, an investment, acquisition, financing or voting decision (see Disclaimer).
- Declared falsifiability. Every verdict is a judgment valid at the as-of date, accompanied by its own falsifier (the observable condition that would disprove it) and by the confidence level.
- Tracked limits. Unverified data is marked [NV]; market data carries source and timestamp [MKT]. The system always declares what could not be verified.
- Professional use. The service is intended for professional users (analysts, investors, advisors), not for consumers or minors.
5. Data, training and confidentiality
- No training on your data. Uploaded documents and outputs are not used to train models. The text of the documents is sent to the model's API only to run the analysis passes requested by the user (see Privacy).
- Ephemeral documents. Uploaded files are processed in memory to extract their text and are not kept as files; the structured analysis remains in the workspace, isolated per-user at the database level.
- Input governance. The quality of the analysis reflects the sources supplied by the user; errors or gaps in the sources propagate to the output within the limits of what the forensic batteries detect and declare.
6. Contact for AI matters
For questions about transparency, human oversight, or to report an output believed to be wrong, use the controller's contact details indicated in the platform or on the invoice. Reports are part of the system's quality and oversight safeguards.