No model training on your data
LiveWe do not train models on customer content. There is no opt-in, because there is no training programme.
- ✓No model training or fine-tuning on customer content
- ✓Three model vendors, all disclosed: Anthropic, OpenAI (embeddings, speech and multi-agent), Google (multi-agent)
- ✓Sub-processor list generated from live configuration, published unauthenticated
- ✓Every model vendor hard-disabled when the deployment is air-gapped
What it is.
Your documents are not used to train or fine-tune any model. That is not a setting you have to find and switch off — there is no training pipeline in the product, and no per-organisation opt-in toggle. LexVio trains no model of its own: the governing-law context an analysis needs is supplied to the model with each request, not learnt from anyone's documents.
The models we send content to are third-party APIs — Anthropic for the main analysis path, OpenAI for embeddings, speech and part of the multi-agent path, and Google for the other part. Their API terms, not their consumer terms, govern that content. The current list is published on our privacy policy page, which is generated from live configuration rather than maintained by hand.
Three steps.
End to end.
There is no training toggle, because there is no training on your content.
Three model vendors receive content on different paths. All three are disclosed and all three are disabled outright in air-gapped deployments.
The sub-processor list is unauthenticated and derived from live configuration — read it rather than take our word for it.
What you get.
- ✓No model training or fine-tuning on customer content
- ✓Three model vendors, all disclosed: Anthropic, OpenAI (embeddings, speech and multi-agent), Google (multi-agent)
- ✓Sub-processor list generated from live configuration, published unauthenticated
- ✓Every model vendor hard-disabled when the deployment is air-gapped
- ✓No per-organisation training toggle — there is nothing for one to control
Quick answers.
No. If we ever build one, it will appear here and in the DPA before it appears in the product.
Error monitoring and product analytics are gated behind a single flag that is off in air-gapped deployments. Model spend and token counts are logged per request against your organisation, which is your data about your usage, not your content.
More in AI Controls & Guardrails.
Every citation comes back with a verdict — verified, ambiguous or unresolved — and a match on title alone is never verified.
Each legal claim we hold about a jurisdiction is checked against that government's own legislation service — and where it cannot be, the answer is "unverified", not silence.
User-pasted text is sanitised before reaching the LLM.
A monthly AI token quota per plan, an 80% warning by email and in-app, and a hard stop at 100%.