AI Policy & Transparency

1. Nature of the AI System

Asystio's AI assistant is powered by large language models (Google Gemini / Vertex AI) that generate responses based on provided information and knowledge bases. Under the EU AI Act, this is classified as a general-purpose AI system with human interaction. While not currently classified as high-risk, we implement appropriate transparency and control measures.

2. AI Disclosure

Every person interacting with Asystio's chatbot is informed they are communicating with an AI system, not a human. This is communicated through:

• Chat interface labels and welcome messages identifying the AI assistant.
• Documentation and help materials clearly stating that an AI handles queries.
• Standard disclosure clauses provided to business clients for their end users.

3. Human Escalation

Asystio includes human-in-the-loop mechanisms. If the AI cannot confidently answer a query, or if the user requests human assistance, the conversation can be escalated to a human operator (when configured by the business client). Critical decisions or sensitive responses can always be reviewed by a human upon request.

4. AI Content Labeling

All AI-generated responses are marked as such. We plan to implement content watermarking to enable automated detection of AI-generated text, in compliance with EU AI Act requirements for synthetic content labeling.

5. Safety and Bias Prevention

• Output moderation: Content filters prevent the AI from generating illegal instructions, hate speech, or offensive content. We monitor logs and user feedback to detect and correct unintended bias.
• Anti-discrimination: We curate training data and knowledge bases to avoid reinforcing stereotypes. Discriminatory outputs are treated as critical bugs requiring immediate correction.
• Scope limitations: Asystio AI is not intended as a sole advisor for medical, legal, or financial decisions. Appropriate disclaimers are displayed to users.

6. User Rights

End users who do not wish to interact with AI can request human assistance. Business clients can disable the AI chatbot and route conversations to staff during business hours. Users are never subject to fully automated decisions without the option of human review.

7. Traceability

We maintain logs of all AI interactions (questions and responses) for 90 days, along with the model version and knowledge base state. This enables investigation of any reported harmful or incorrect responses and supports audit requirements.

8. Error Reporting

Users can rate AI responses (thumbs up/down) and report issues. Business clients can contact [email protected] for AI-related incidents. AI errors are investigated promptly with corrective actions (knowledge base updates, prompt adjustments).

9. EU AI Act Compliance

• Risk assessment: We conduct regular internal AI risk assessments to identify and mitigate potential negative impacts.
• Human oversight: Asystio AI is a support tool - all binding decisions remain with human managers.
• Data quality: As we develop fine-tuned models, we document data provenance and test for bias.
• Registration: If required by regulation, we will register our AI system in the official EU registry.

10. AI Ethics

• AI always identifies itself as AI - it never impersonates a human.
• AI strives for impartiality and encourages consulting human experts for complex decisions.
• Data used by AI is minimized to what is necessary.
• We are open to external audits and advisory input on AI ethics.