How SovEU uses AI

AI in SovEU advises; a person decides. The model never sets a compliance value, a deadline or a risk class, never closes a case, and never files a record on its own. Those outcomes are computed by deterministic rules or chosen by a human — the AI only drafts, extracts and retrieves, and every output is reviewed before it is saved or acted on.

SovEU's AI assist — “Ask SovEU” — does three things:

• Drafting — first drafts of notices, responses and summaries that you edit and approve.
• Extraction — pulling structured details out of documents you upload, for you to confirm.
• Search — embeddings that power document and knowledge search, with answers that cite their sources.

AI assist runs on Mistral (France) — an EU model provider — not a US model. We previously used Anthropic for this and have since moved fully to Mistral. The full list of third parties we rely on is on our public sub-processor register.

When you invoke an AI feature, only the content needed for that task is sent to the provider — the document you are extracting, the text you are drafting from. We use the provider's API under data-processing terms that prohibit using your data to train their models. Your content is not sold, not used for advertising, and not shared beyond what the feature you triggered requires.

Every AI output is reviewed by a person before it is saved or acted on. AI assist is feature-gated and optional — you can turn it off per workspace and run SovEU entirely without it. Turning it off never removes a compliance capability, because compliance values, deadlines and risk classes are produced by deterministic logic, not the AI.

AI requests are sent to perform the task and are not retained by us beyond it. Search embeddings are stored inside your own workspace and are removed when you delete the underlying content or your data. See our privacy policy and security overview.

Ask us anything about our AI use or request our data-processing terms at privacy@soveu.io.