Where Your Client Data Goes When AI Drafts

When the report engine drafts a section, it sends that engagement's findings and scope to an AI model and receives text back. Nothing about your client's data is retained by the model itself. Inference is stateless: the request goes in, the draft comes out, and there is no memory of it afterwards.

Because the AI runs on your own key, the account is yours and so is the agreement governing it. Retention settings, data-processing terms and any regional commitments are between you and your provider, not mediated by us. That is deliberate: you can produce that agreement in your own audit without asking us for anything.

What this means in practice

• Your engagement data is not training data. On the major providers' API terms, prompts sent on your own key are not used to train models, and because the account is yours you can verify that yourself.
• The model does not retain your data. Each request is independent, so nothing from one engagement can surface in another.
• Your usage is never pooled with another firm's. Separate key, separate account, separate boundary.
• Every drafted section is reviewed by your operators before it reaches a client.
• Drafting spend is metered per engagement, so you always see what it cost.