Disclose AI involvement
Art. 50(1)Users must be informed when they're interacting with an AI system. APEX surfaces an 'AI-generated' label on every response, with the model name, version, and run timestamp visible in the citation panel.
EU AI Act · Article 50
Article 50 ready by construction.
EU AI Act Article 50 took effect August 2, 2026. Every PYRAMYD answer ships with the per-claim provenance, machine-readable AI-content marking, and audit chain procurement teams now require · not as an add-on, but as the substrate.
Effective Aug 2, 2026Up to €15M / 3% turnover penalty100% AI responses labeledPer-claim citation chain
The four transparency obligations
What Article 50 requires · and what PYRAMYD does about each.
The four sub-articles map to concrete UX and audit-log behaviors. Below is how PYRAMYD satisfies each.
Mark generated content as machine-readable
Art. 50(2)Synthetic content must carry a machine-detectable marker. Every APEX response embeds a structured-data block (JSON-LD) with the AI flag, content hash, and generation metadata · scrapeable by downstream auditors without changing the visible UX.
Label emotion-recognition and biometric categorization
Art. 50(3)Not applicable today · PYRAMYD does not run emotion recognition or biometric categorization. Disclosure remains in the model card for completeness.
Label deepfakes / artificial content
Art. 50(4)AI-generated text, images, audio, or video must be labeled. APEX outputs are labeled at the response level; exported assets (PDF, slides, sheets) carry the same provenance block in the file metadata.
Provenance schema
Every cell carries eight pieces of evidence.
Click any field in the workspace and the provenance popover shows the complete chain. The same schema is exported with every PDF, slide, sheet, or API response.
| Field | What it carries |
|---|---|
| Source URL | Every claim links to its origin (G2 review, SEC filing, press release, etc.) |
| Retrieval timestamp | When PYRAMYD fetched the source · separate from when the answer was generated |
| Model name + version | Claude Opus 4.7 · GPT-5 · Gemini 2.5 Pro · per-call, not platform-level |
| Prompt hash | SHA-256 of the prompt template + variable bindings · stable across reruns |
| Quality score (0-100) | Signal-strength score from the verification gate, not a star rating |
| Confidence score (0-100) | Model's self-reported certainty on the extraction · separate from quality |
| Citation set | Ordered list of sources that supported each claim · multi-hop trails resolved |
| Audit log entry ID | FK back to the per-answer log row · who asked, when, what was retrieved |
Timeline
Key Article 50 dates.
- July 12, 2024Regulation (EU) 2024/1689 (EU AI Act) published in the Official Journal
- August 1, 2024EU AI Act entered into force
- February 2, 2025Prohibitions on certain AI systems became applicable (Art. 113)
- August 2, 2026Article 50 transparency obligations became generally applicable (Art. 113)
Source dates per Regulation (EU) 2024/1689, Article 113 (Entry into force and application) and the EU AI Act application timeline published at eur-lex.europa.eu.
Related compliance frameworks
The standards Article 50 sits inside.
EU AI Act Article 50 doesn't stand alone · it sits on top of a stack of AI-management and AI-risk frameworks. PYRAMYD's provenance schema is compatible with all three.
ISO/IEC 42001:2023
First international standard for AI Management Systems (AIMS). Defines how organizations establish, implement, maintain, and continually improve an AI management system. Published Dec 2023 by ISO/IEC.
NIST AI Risk Management Framework 1.0
Voluntary US framework (NIST AI 100-1, January 2023) for managing AI risks. Pairs with the GenAI Profile (NIST AI 600-1, July 2024). Provenance and traceability are explicit AI RMF characteristics.
Regulation (EU) 2024/1689
The EU AI Act itself · official text at eur-lex.europa.eu/eli/reg/2024/1689/oj. Article 50 governs transparency; Article 99 sets penalties (up to €15M or 3% of worldwide annual turnover, whichever is higher).
For procurement teams
The one-line answer to every Article 50 RFP question.
“Every PYRAMYD response carries an AI-disclosure flag, a machine-readable provenance block (JSON-LD), per-claim citation chain, model + prompt + timestamp, and a queryable audit log retained for the lifetime of your contract plus seven years · aligned with Article 50(1), 50(2), and 50(4) of the EU AI Act and NIST AI RMF 1.0 provenance characteristics.”
Paste into any RFP response. We'll back it up with the technical evidence on the call.
Walk your compliance team through the audit chain.
30 minutes. We show the per-cell provenance, the JSON-LD content marking, the audit-log export, and the seven-year retention contract terms. Send your compliance officer.