How does Unloq use AI models with your data?

Which AI models does Unloq use?

Unloq is model-agnostic and sends each task to a model that fits it. Unloq uses frontier models, Anthropic’s Claude among them, and models your own data scientists have built can plug in. Each answer’s trail shows the plan Unloq made and the SQL that ran, so a reviewer can check the work whichever model did it.

Is our data moved out to train models?

When Unloq runs in your cloud, your data is never moved out to train anything. Your data is read to answer your team’s questions, and each answer shows the sources it used.

How does Unloq check what the model produces?

Figures are computed in code from your data and checked against their source before you see them, and Unloq would rather show nothing than a confident wrong answer. A metric resolves to one canonical, team-owned definition before any query runs, and the model only plans while deterministic code executes. Any answer that uses a document cites it.

When your data or documents do not support an answer, Unloq says there is not enough evidence, and where the data is partial, Unloq says so. An answer that breaks a business rule you have agreed is blocked before anyone sees it. The trail on each answer holds the plan Unloq made, the SQL that ran, the documents with the page and excerpt for each, and a validation grade, and how every answer is evidenced walks through each part.

Who in your organisation can see the answers?

Access is by role: a CEO sees everything, while an account executive sees what their role allows. Role-based access is checked at the moment each query runs, and people sign in with single sign-on through your existing identity provider (SSO / OIDC). A regional executive at a global utilities group made the point to us in July that who can see commercially sensitive data matters even when hosting is inside the company’s own environment.

Can an answer draw on a folder someone’s role has not been granted?

No. If a role has not been granted a folder Unloq can read, the role check stops that role’s questions pulling from it. A folder Unloq is never connected to cannot appear in an answer at all, so what your team connects decides what an answer can draw on. Any answer that uses a document cites the exact document and page, so a reviewer can see what it drew on.

What does Unloq do on its own?

Once your team connects a folder, Unloq reads and indexes the documents in it, so answers can cite the exact document and page. Unloq keeps watch on your numbers and tells whoever owns the one that moved. Unloq reads the data you approve and does not change or delete it. Read-only applies in both options, whether Unloq runs in your cloud or we host it for you. Answers reach named owners in Unloq, your CRM or Slack, and a person decides what, if anything, changes in your systems as a result. A recommendation that would breach a limit you have set is flagged before you approve it. A person signs off every decision, and Unloq records the approval with the outcome you expected.

What this page does not cover

The four ways in are on how Unloq connects to your data, and the deployment options on where your data sits. Certifications, and a table for your reviewer’s own sheet, are on what a security review needs from Unloq. The platform overview sets out the four layers, from what happened to what it was worth, and the security pack request is where you tell us what your review needs.

Ask us your question

If your AI governance review has a question this page leaves open, ask us your question. Our AI engineering team works the question through with our technical founders and replies with an answer.