AI Ethics Policy
AI features in health and wellbeing products carry significant ethical responsibilities. This policy sets out the principles that govern every AI feature across all Viverio Ltd products, current and future.
1. Our AI Principles
Viverio Ltd uses AI as a tool to make evidence-based health support more accessible — not as a replacement for human clinical judgement, professional care, or crisis services. Every AI feature we build across all products is governed by six principles:
| Principle | What it means in practice |
|---|---|
| Safety first | AI features have hardcoded crisis overrides. When risk content is detected, the AI does not respond with therapy — it returns crisis resources only. This rule cannot be overridden by any user, system prompt modification, or model instruction. |
| Honest labelling | All AI-generated content is prominently labelled as AI-generated and not clinical advice. We never present AI output as human-authored, clinically reviewed, or therapeutically authoritative. |
| Explicit consent | AI features require separate, explicit, revocable consent before first use. Users who do not consent to AI features have full access to all other product functionality. |
| Privacy by design | Only the specific text typed for an AI feature is transmitted to our AI provider. No health history, mood data, journal entries, or other app data is ever sent to the AI provider. |
| Scope limitation | AI features in Viverio Ltd products are limited to reflection, journalling, and psychoeducational support. AI features do not perform symptom assessment, diagnosis, prescribing, or clinical treatment planning. |
| Human oversight | AI system prompts are reviewed at every material product release. The Clinical Safety Officer reviews AI safety controls as part of the pre-release governance checklist. |
2. AI Safety Controls
The following safety controls apply to all AI features across all Viverio Ltd products:
- Crisis override: a hardcoded system prompt rule that detects risk content and returns crisis resources only — bypassing all other AI behaviour
- Client-side crisis detection: keyword detection fires before content reaches the API, preventing harmful content from being sent
- Moderate-risk content handling: content indicating distress but below crisis threshold triggers compassionate framing and mandatory crisis signposting
- Token limits: AI responses are capped to prevent extended quasi-clinical reasoning
- Server-side proxy: API keys are never embedded in app binaries
- System prompt prohibition: AI system prompts explicitly prohibit diagnosis, prescribing, and clinical recommendation language
3. What Our AI Does Not Do
Viverio Ltd AI features do not diagnose any medical or psychiatric condition. They do not recommend medications or treatments. They do not replace clinical assessment, therapy, or crisis services. They are not supervised by mental health professionals. They are reflection and journalling aids only.
4. AI Provider Selection and Oversight
Viverio Ltd currently uses Anthropic’s Claude models as its AI provider. When selecting or changing AI providers across any product, Viverio Ltd considers: the provider’s published safety and ethics commitments; data processing terms; compliance with UK and EU data protection law (including SCCs); and the provider’s crisis and harmful content handling capabilities. A Data Processing Agreement (DPA) is required with all AI providers before any user data is processed.
5. Transparency with Users
- AI features are described plainly in the product FAQ of each application
- The AI consent modal explains what is sent to the AI, what is not, and how to revoke consent
- All AI-generated content is labelled at the point of display
- The Privacy Policy and product DPIAs describe AI data flows in plain language
6. Bias and Fairness
AI models can produce biased, culturally inappropriate, or inaccurate outputs. Viverio Ltd mitigates this risk by: specifying UK English, inclusive, and non-clinical language in all system prompts; reviewing AI output quality as part of pre-release testing; providing users with a feedback mechanism for inappropriate AI responses; and monitoring complaints for patterns suggesting systematic AI output issues.
7. Review and Accountability
This policy is reviewed annually and whenever a material change is made to AI features or the AI provider across any Viverio Ltd product. The Director is accountable for compliance with this policy. The CSO reviews AI safety controls as part of clinical governance sign-off for each product.