EU AI Act Compliance
Transparency documentation in accordance with Regulation (EU) 2024/1689 (EU AI Act)
1. AI System Classification
nexAI.BIMcheck is classified as a Limited-Risk AI Assistant System in accordance with Article 50 of the EU AI Act (Transparency obligations for certain AI systems).
Justification: nexAI.space is a support tool that assists engineers, architects, and BIM coordinators in quality assurance of building information models. The system does not make autonomous decisions – the final decision always remains with the qualified professional. It does not fall under Annex III (High-Risk AI) as it does not make autonomous safety-critical decisions nor functions as a safety component of a product.
Risk Level: Limited Risk (Article 50)
Application Domain: Construction / BIM Quality Assurance
Decision Type: Supportive only (no autonomous decisions – final decision always with the engineer)
Why not a High-Risk System?
Annex III of the EU AI Act lists high-risk areas such as critical infrastructure, education, employment, law enforcement, or biometric identification. nexAI.space analyzes BIM models for technical errors and inconsistencies – comparable to an automated code analysis or spell-checker. The results are suggestions, not binding decisions. Full responsibility for construction planning remains with the qualified professional.
2. AI Models Used
nexAI.BIMcheck uses the following AI models:
| Model | Purpose | Provider |
|---|---|---|
| Mistral AI 70B | Anomaly detection, model analysis | Mistral AI (EU) |
| Fine-tuned models | IFC analysis, rule checking | Proprietary (EU servers) |
| Ensemble methods | Increased reliability | Proprietary (EU servers) |
Training Data and Anonymization
All data is fully anonymized before use in training. Since only anonymous data is processed, the GDPR does not apply to the training process (Recital 26 GDPR).
What is removed (before training):
- User IDs, project IDs, project names
- Building names, addresses, GPS coordinates
- File names, author fields, organization metadata
- All GlobalIds (replaced with generic identifiers)
- Timestamps (rounded to month/year)
What is used for training (anonymous):
- Error types and severity levels (e.g. "missing fire rating", "critical")
- Building element classes (e.g. IfcWall, IfcDoor, IfcSlab)
- AI confidence scores and their correctness
- User feedback (correct/incorrect – without user reference)
- Generic building typology (residential, commercial, industrial)
- Geometric patterns and element relationships
3. Transparency Obligations (Article 50)
As a limited-risk AI system, nexAI.space is subject to the transparency obligations under Article 50 of the EU AI Act. These require the provider to inform users that they are interacting with an AI system.
- Clear Marking: All AI-generated results are clearly identified as such
- Confidence Scores: Each analysis result includes a confidence score (0–100%)
- Explainability: Detailed reasoning is provided for each finding, including references to applicable standards
- Limitations Disclosure: Known limitations of the AI system are documented and communicated to users
- Recommendation Character: Results are explicitly presented as recommendations, not binding statements
4. Human Oversight (Design Decision)
nexAI.space is designed as an assistant system. Final decision authority always lies with the qualified professional. This is not a regulatory requirement for limited-risk systems, but a deliberate design choice.
- No Autonomous Decisions: The AI system provides recommendations only – all decisions must be made by qualified professionals
- Human Verification Required: Critical findings require human review before action is taken
- Override Capability: Users can dismiss, modify, or override any AI recommendation
- Qualification Requirements: The system is designed for use by qualified construction professionals (architects, engineers, BIM coordinators)
5. Voluntary Quality Measures
Beyond the legally required transparency obligations, nexAI.space voluntarily implements additional quality measures. These are not legally required for limited-risk systems, but reflect our commitment to quality.
- Continuous Monitoring: All AI outputs are logged and analyzed for anomalies
- Bias Testing: Regular bias audits across different building types, sizes, and software tools
- Performance Metrics: Accuracy 94.2%, Precision 92.8%, Recall 95.6%, F1-Score 0.942
- Version Control: All model updates are versioned and can be rolled back
- User Feedback: Continuous collection and analysis of user feedback
6. Data Governance and Privacy
nexAI.space processes all user data in accordance with the GDPR:
- European Servers Only: All data processing on IONOS servers in Germany
- No Third-Country Transfer for AI Processing: All AI models and BIM data processing run exclusively on EU servers. Supplementary services (CDN, CRM) operate under EU Standard Contractual Clauses – see Privacy Policy
- Automatic Anonymization: All data is fully anonymized before ML training – no inference to projects or persons possible
- Data Minimization: Only technically necessary data is processed
- Data Sovereignty: Users retain full control over their data
7. Transparency Declaration (Article 50)
zeit+raum (Alexander Maier) hereby declares that the AI system nexAI.space complies with the transparency requirements of Regulation (EU) 2024/1689 for limited-risk AI systems:
Legal obligations (Limited Risk):
- Article 50 – Transparency: ✓ Users are informed they interact with AI
- Article 50 – Marking: ✓ AI-generated content is marked as such
- Article 50 – Documentation: ✓ This page documents the AI systems and their operation
Not applicable (not a high-risk system):
- Article 9 – Risk management system: Not required (voluntarily partially implemented)
- Article 10 – Data governance: Not required (voluntarily: anonymization documented)
- Article 11 – Technical documentation: Not required (voluntarily: this page)
- Article 17 – Quality management system: Not required (voluntarily partially implemented)
- Article 49 – Conformity assessment: Not required (high-risk systems only)
Prohibited practices (Article 5):
- No manipulative or deceptive techniques ✓
- No exploitation of vulnerabilities ✓
- No social scoring ✓
- No biometric identification ✓
Declaration Date: August 2026
Next Review: February 2027
Responsible: Alexander Maier, zeit+raum
8. Contact for AI Compliance
For questions regarding our AI systems, their compliance with the EU AI Act, or to report incidents:
zeit+raum
Alexander Maier
Elbestraße 11, 55122 Mainz, Germany
Email: info@zeitundraum.de
Phone: +49 (0)6131 – 32 71 400
Supervisory Authority
Bundesnetzagentur (German Federal Network Agency)
As designated market surveillance authority for AI systems in Germany
Last updated: August 4, 2026