AI Transparency

1. What AI Model We Use

Vetta uses Anthropic Claude, a large language model (LLM) developed by Anthropic. Claude is trained on a broad range of text data, including veterinary literature, medical journals, and educational resources about animal health.

The Claude model is specialized to provide educational information about dog health topics in a conversational manner. It does not provide veterinary diagnosis or medical treatment recommendations.

2. What the AI Does and Doesn't Do

The AI Does:

• Provide general educational information about dog breeds, health conditions, and wellness
• Help users understand health concepts and ask better questions for their veterinarian
• Suggest common preventative care measures and wellness practices
• Explain veterinary terminology and procedures
• Help organize and interpret health records

The AI Does NOT:

• Provide veterinary diagnosis or medical treatment plans
• Replace professional veterinary consultation
• Make medical decisions for your dog
• Prescribe medications or treatments
• Provide emergency medical advice
• Guarantee accuracy of all health information

3. How Conversations Are Processed

When you send a message to Vetta:

1. Your message is sent to Anthropic's servers: Your chat message is transmitted to Anthropic's API servers (located in the United States) for processing by the Claude model.
2. AI generates a response: The Claude model processes your message and generates an educational response based on its training and the context of your conversation.
3. Response is returned: The generated response is sent back to your device and displayed in the chat interface.
4. Conversation is stored: Both your message and the AI response are stored in Vetta's database (hosted on Supabase in the EU) for record-keeping and service continuity.
5. AI provider data handling: Anthropic retains API logs for up to 30 days for safety monitoring, then deletes them automatically. Anthropic does not use your conversations to train its AI models.

Please review our Privacy Policy for details about data processing and international transfers.

4. How Internal Analysis Works

Vetta may analyze your conversations internally for:

• Understanding patterns in your dog's health (e.g., recurring symptoms)
• Providing continuity in your health records across conversations
• Improving our Service and AI responses
• Detecting potential safety issues or emergency situations

Important: This internal analysis is never displayed to you and is used only for the purposes listed above. The analysis results do not influence the educational responses you receive.

5. Safety Measures

Vetta implements multiple safeguards to ensure safe and appropriate AI responses:

Multi-Layer Filtering (4 Layers)

1. Input Filtering: User messages are analyzed to detect inappropriate content or requests that fall outside the scope of pet health education.
2. Context Awareness: The system maintains awareness of conversation context and prevents the AI from making medical claims or recommendations.
3. Output Filtering: AI-generated responses are analyzed before display to remove any medical claims, diagnoses, or treatment recommendations that slipped through.
4. Post-Generation Review: Emergency or health-critical responses are flagged for human review.

Emergency Detection

Vetta includes systems to detect potential emergency situations (e.g., mentions of severe symptoms, trauma, or poisoning) and automatically alerts users to seek immediate veterinary care.

Disclaimer Consistency

All responses include clear language that Vetta is educational and not a substitute for professional veterinary care.

6. Human Oversight

Vetta includes systems for human oversight:

• User Feedback System: Users can rate AI responses and report concerning or inappropriate behavior. Flagged conversations are reviewed by our team.
• Regular Audits: We periodically review AI responses and user interactions to ensure safety and appropriateness.
• Continuous Monitoring: System logs are monitored for unusual patterns or potential safety issues.

7. Limitations and Known Risks

While we strive for accuracy, users should be aware of limitations:

• Hallucinations: AI models can sometimes generate plausible-sounding but incorrect information. Always verify important health information with a veterinarian.
• Outdated Information: Training data has a knowledge cutoff and may not reflect the latest veterinary research.
• Lack of Individual Context: The AI cannot perform physical examinations or understand the full medical context of your dog's situation.
• Individual Variability: Each dog is unique; information that applies to one dog may not apply to another.
• No Real-Time Updates: The AI does not have access to current veterinary guidelines or breaking news in animal health.

8. How to Report Issues

If you encounter:

• Inappropriate or harmful AI responses
• Inaccurate health information
• Technical errors or bugs
• Privacy or security concerns

Please contact us at privacy@vetta.health. We take all reports seriously and will investigate promptly.

9. Continuous Improvement

Vetta is continuously improved through:

• Regular updates to safety filters and guardrails
• User feedback and reported issues
• Monitoring of AI response quality
• Integration of latest veterinary knowledge
• Updates to comply with evolving AI regulations

10. Your Rights and Control

You have the right to:

• Understand that you are interacting with AI (disclosed in every chat)
• Request deletion of your conversation data (see Privacy Policy)
• Export your health records at any time
• Report concerning AI responses
• Opt out of optional data analysis for improvement

You retain full control over your data and can request access, correction, or deletion at any time by contacting privacy@vetta.health.

Questions?

If you have questions about our AI systems or transparency practices, please contact:

Email: privacy@vetta.health
Company: Vetta BV, Belgium