Skip to main content
Trusted Office
Sign in

AI use notice

Last updated 17 November 2026.

Trusted Office uses AI to prepare drafts. This notice explains how drafts are made, what checks they go through, and what they are and are not for.

How a draft is made

  1. Someone in your organization starts a playbook and answers its questions.
  2. Trusted Office finds the passages in your organization's documents that matter for the task.
  3. Those passages, your answers and the playbook's instructions are sent to an AI model, which writes the draft. Today the drafts are written by Anthropic's Claude models through OpenRouter, and the search index is made with Google's embedding model.
  4. Our software checks the draft (see below) and shows it to the people who must approve it.

The AI model gets text only. It cannot send email, change files, or take any other action itself.

Checks on every draft

  • Sources are checked. When a draft quotes or relies on your documents, each quote must match the source exactly. Claims we cannot match to a source are removed and marked "Removed: could not verify a source."
  • Sensitive information is blocked. Uploads, run answers and drafts are checked for card numbers, bank details, government ID numbers, passwords and keys, and health and student records.
  • Limits are enforced. Playbooks that write to people outside your organization have a maximum number of recipients and items, set by your administrators.
  • Every draft is labelled "Draft — review before using".

What approvals do

Nothing leaves Trusted Office until a person approves it. The approver sees the draft next to its sources and a checklist of what a good result looks like, can edit it, and decides whether to approve. Each approval is recorded in your organization's activity log with the approver's name and time. Approved email drafts are saved in the approver's own Gmail, if connected, or in Trusted Office; they are never sent automatically. In one-person organizations on the Free and Starter plans the same person may approve their own drafts; those approvals are labelled "self-approved".

Your data and AI training

We never use your documents, answers or drafts to train AI models. Every request to the AI service asks for zero data retention and turns off data collection, so the model provider may not keep or train on your content.

The limits of drafts

AI can be wrong. A draft may misread a document, miss something important, or sound more certain than it should. Drafts are a starting point for your own work. They are not legal, tax, financial, medical or employment advice, and they should not be the only basis for decisions that seriously affect a person, such as hiring, firing, credit, housing or care. When it matters, check with a qualified professional.

Questions and problems

If a draft is wrong in a way that worries you, or you think the AI behaved unexpectedly, tell us at hello@trustedoffice.app. Security problems go to security@trustedoffice.app.

We use essential cookies to make the site work. With your permission we also count visits with our own analytics, never shared with anyone else. Privacy policy