EU AI Act — Classification Notice (Caelex products)

1. Scope — which products

Caelex operates several software products with generative and assisting AI features. The following classification applies across products, with product-specific transparency implementations:

• Atlas — research, drafting and mandate administration for professionals (space/reg)
• Passage (Trade) — export control, screening (CIS), classification including vision/LLM extract
• Comply Generate 2.0 — AI-assisted first draft of compliance documentation
• Scholar — pedagogical research and simulations (space law)
• Aether / Ephemeris — operational surfaces with limited AI assistance, where implemented
• Public website — marketing; no decision-making AI for third parties

2. Common role of the products

Caelex products are AI-assisted tools in support of professionals and compliance teams. They do not take autonomous decisions with legal effect vis-à-vis authorities or courts; the human remains the decision-maker. Outputs are drafts, screening indications, classification proposals or research aids for independent review.

3. High-risk (Annex III)? — No, on our analysis

Annex III point 8(a) covers, among other things, AI in support of judicial authorities. Caelex products are directed at private professionals and undertakings, not at courts or public prosecutors as the intended purpose, and they do not take judicial or administrative decisions.

• Use by attorneys and compliance teams, not by judicial authorities as the intended purpose
• Outputs are drafts, proposals and indications for independent review
• No automatic filing or clearance vis-à-vis authorities by Caelex software
• Passage CIS and ship-gates enforce human clearance (fail-closed) — no autonomous export clearance

4. Transparency and limited risk (Art. 50)

On our assessment the products are subject to transparency obligations for generative AI / chat systems and AI-assisted extraction. Implementation in the product includes:

• Labelling as an AI-assisted system (product branding, UI notices)
• Notices that outputs are not legal advice and must be reviewed before use
• Atlas and Comply Generate 2.0: server-side disclaimer inject on export/complete (Art. 50 parity)
• Passage Vision/Extract: UI notices that extraction is AI-assisted and must be verified manually
• Human control: accept, discard, edit; write actions with confirmation, where implemented
• Public memo page: /legal/eu-ai-act-en; AI transparency: /legal/ai-disclosure-en

5. Data quality and hallucination mitigation

• Curated corpora and control lists with source IDs and validity status, where wired into the product
• Citation, screening and classification tools for traceability
• Fail-closed ship-gates and golden-corpus tests on safety-critical paths (Passage CIS)
• No claim of completeness or error-free AI output

6. Underlying models

Where generative inference is used, Caelex uses Anthropic Claude EU-only via AWS Bedrock in the EU (AI Gateway, fail-closed). Governance and certification statements of the provider are in the Anthropic/AWS documentation. Caelex controls routing and product disclaimers; residual risks of model inference remain. Caelex does not claim its own model certification.

7. Conformity assessment

On our classification as not high-risk, no external conformity assessment under Art. 43 is required. Caelex documents the classification internally and reviews it upon material product changes. Formulations such as “AI Act certified” or “fully certified as compliant” are not used in marketing or product.

8. Timeline (orientation)

• Regulation (EU) 2024/1689: entry into force 1 August 2024
• Phased applicability (prohibitions, GPAI, high-risk) per the statutory deadlines
• Caelex products: focus on transparency and human oversight; high-risk duties are not triggered on the current classification