This document explains how RisqBase d.o.o. designs, operates, and governs AI systems within RALIA. We are committed to responsible AI development that aligns with our values and applicable regulations, including the EU AI Act.
1. Our AI Principles
Our approach to AI is guided by five core principles:
Transparency
We are open about how our AI works, its capabilities, and limitations. Users can always distinguish AI-generated content from human input.
Human Oversight
AI provides assistance, but humans remain in control of all meaningful decisions. Our systems augment, not replace, human judgement.
Privacy by Design
We minimise data collection, do not use customer data to train models, and implement robust security measures.
Accuracy & Reliability
We continuously improve AI accuracy and clearly communicate when outputs should be verified by professionals.
Fairness
We design systems to treat all users fairly and work to identify and mitigate potential biases in AI outputs.
2. How We Use AI
RALIA incorporates AI technology across four key areas:
LIA Assistant
Our AI-powered Legal Intelligence Assistant helps you navigate compliance requirements:
- Answering questions about regulatory frameworks
- Providing guidance on AI Act compliance
- Explaining technical concepts in plain language
- Suggesting relevant documentation and resources
Assessments
AI helps users complete compliance assessments by:
- Providing context-aware question explanations
- Suggesting appropriate responses based on input
- Generating preliminary risk classifications
- Identifying potential compliance gaps
Documents
AI assists in creating compliance documentation:
- Fundamental Rights Impact Assessments (FRIAs)
- Data Protection Impact Assessments (DPIAs)
- Transparency documentation
- Risk assessment reports
Horizon Intelligence
AI enhances our regulatory monitoring service:
- Summarising regulatory updates and guidance
- Identifying relevant content based on your preferences
- Analysing trends in regulatory developments
- Generating digest emails and alerts
3. AI Governance
3.1 Risk Classification
Limited Risk AI System
Under the EU AI Act, RALIA is classified as a limited risk AI system. Our primary obligation is transparency about AI use, fulfilled through clear labelling and this documentation.
RALIA is not used for and explicitly prohibits use cases that would constitute high-risk or prohibited AI applications under the EU AI Act.
3.2 Oversight & Accountability
- Regular review of AI outputs for quality and accuracy
- Monitoring for potential bias or harmful outputs
- Clear escalation procedures for AI-related issues
- Designated personnel responsible for AI governance
3.3 Model Selection & Providers
We carefully select and evaluate the AI models that power RALIA:
| Provider | Role | Data Protection |
|---|---|---|
| Anthropic (Claude) | Primary | EU Data Processing Agreement |
| OpenAI | Backup | EU Data Processing Agreement |
All providers are contractually bound to data protection requirements with regular evaluation of performance and safety.
4. Data & Privacy
Your Data Is Not Used for Training
We do NOT use customer data to train our AI models. Your compliance information, assessment responses, and documents are processed to provide results, but are never used to train or improve the underlying AI systems.
Data Responsibility
RALIA
- Encrypts all data in transit (TLS 1.3)
- Providers do not retain data beyond session
- Maintains audit logs for security
- Minimises data sent to AI providers
You
- Choose what information to share
- Review AI outputs before use
- Avoid inputting sensitive PII
- Report any data concerns
5. Transparency & Labelling
5.1 AI Content Identification
All AI-generated content in RALIA is clearly identified:
LIA Responses
Labelled as AI-generated
Documents
Include AI disclaimers
Recommendations
Indicate AI involvement
Summaries
Note their AI origin
5.2 Limitations Disclosure
Important Limitations
- • AI outputs may contain errors or inaccuracies
- • AI guidance does not constitute legal advice
- • Complex situations require professional human review
- • AI knowledge has temporal limitations (training cutoff dates)
6. Human Oversight
User Control
- AI features can be used optionally
- Edit, override, or reject AI suggestions
- Final decisions always rest with you
- Request human support at any time
Review Processes
- Risk classifications are marked as preliminary
- Generated documents are presented as drafts
- Recommendations include professional verification caveats
Escalation
Escalate concerns about AI outputs through our support channels. All escalations are reviewed by qualified human staff.
7. Safety & Security
7.1 Content Filtering
Our AI systems include safeguards against:
7.2 Security Measures
TLS 1.3
Secure API communications
Rate Limiting
Abuse prevention
Access Controls
Authentication required
Monitoring
Pattern detection
7.3 Incident Response
- Rapid assessment of potential harms
- Temporary suspension of features if necessary
- User notification where appropriate
- Root cause analysis and remediation
8. Continuous Improvement
8.1 Monitoring & Updates
We continuously monitor and improve AI performance:
8.2 Regulatory Compliance
We monitor evolving AI regulations and maintain compliance with:
- EU AI Act requirements
- GDPR automated decision-making obligations
- Industry standards and best practices
- Regulatory authority guidance
9. Your Responsibilities
As a user of RALIA's AI features, you are responsible for:
- 1Using AI features for their intended purpose (compliance assistance)
- 2Reviewing AI outputs before making decisions
- 3Seeking professional advice for complex matters
- 4Reporting concerning, inaccurate, or harmful outputs
- 5Complying with our Acceptable Use Policy
- 6Not attempting to manipulate or abuse AI systems
10. Contact & Feedback
We welcome feedback about our AI systems and this documentation.
General Enquiries
support@risqbase.com
AI Concerns
ai-questions@risqbase.com
Privacy
privacy@risqbase.com
Use the feedback buttons within the platform or contact us directly to report issues.
Document History
| Version | Date | Summary |
|---|---|---|
| 2.0 | January 2026 | Enhanced interactive design, improved navigation |
| 1.0 | January 2026 | Initial publication |