Trust Centre
Responsible AI
Last updated August 2026 · v1.4 · Changelog
A9.1 How VIQ Works
VIQ is a retrieval-augmented generation (RAG) system. User queries trigger semantic retrieval from Vixio's proprietary regulatory corpus via Pinecone vector search. Retrieved content is passed as context to a large language model (currently from Anthropic's Claude family, accessed via AWS Bedrock) to generate a response. The regulatory corpus, maintained by Vixio's editorial and analyst team, is Vixio's primary intellectual asset. VIQ does not access the open internet.
A9.2 Model Governance
Vixio operates a continuous model evaluation programme. The specific models deployed may change as Vixio evaluates and optimises across model releases. Model version changes within an approved sub-processor do not constitute the addition of a new sub-processor.
| Model family | Provider | Status |
|---|---|---|
| Claude | Anthropic (via AWS Bedrock) | Primary, in production |
| GPT | OpenAI (via Azure) | Under evaluation |
| Gemini | Google (via Vertex AI) | Under evaluation |
| Llama | Meta | Under evaluation |
| Mistral | Mistral AI | Under evaluation |
| Nova / Titan | Amazon (via AWS Bedrock) | Under evaluation |
A9.3 Human Oversight
Vixio's editorial and analyst team provides human oversight of the regulatory corpus underpinning VIQ. VIQ outputs are not reviewed individually prior to delivery to subscribers. Subscribers are responsible for verifying the accuracy and applicability of outputs as set out in A6.3. Vixio does not use AI to make automated decisions that produce legal effects on subscribers.
A9.4 AI Operating Principles
Customer data, including prompts and document content, is not used to train foundation models for the benefit of other customers.
AI outputs are supplementary to, not a replacement for, professional regulatory and legal advice.
Vixio does not use AI to make automated decisions producing legal effects on or significantly affecting subscribers.
AI sub-processors are contractually restricted from training or improving models on customer data without explicit authorisation.
Material changes to AI infrastructure affecting data processing will be notified to subscribers with not less than 30 days' notice.
The AI sub-processor list and foundation model list are maintained at trust.vixio.com and updated when changes occur.
The contractual terms governing AI Functionality, ownership of Outputs, and your obligation to verify Outputs before relying on them are set out at Section 7 of the Acceptable Use Policy.