The short version
- Nobody trains an AI model on your content. Not us, and not the AI providers we use.
- What goes in and what comes out are yours.
- AI can be wrong. Check what it writes before you rely on it or publish it.
- When you connect your own AI app or key, your provider’s terms apply to what you send it. You can disconnect at any time.
1. Where AI is used
Which features your plan includes is shown in the app. Some features are switched on gradually and may not be available on your account yet.
2. What each one sees
Henry sees your published brand pages, the descriptions of your files, what you are looking at when you ask, anything you attach, and the conversation so far. That is sent to OpenAI or Google, whichever your key belongs to, each time you ask. File descriptions send each image or document you upload to Anthropic once, to write its description. Oswald sees your brand pages, the files you point him to, anything you attach, and, when you ask him to start from your website, the public pages of that site, which our own server reads. What you and he say is sent to Anthropic. A connected AI app sees whatever it asks Brand Atlas for, within the permission you gave it: everything in your brand on Read, and the ability to draft, publish, create Horizons and upload files on Read and write.3. Our AI providers
Anthropic, the provider we pay for, is listed on the Subprocessors page. Under its commercial terms, Anthropic does not use data sent through its API to train its models. [TEAM TO CONFIRM: these terms against our agreement with Anthropic, and how long Anthropic keeps inputs and outputs.] OpenAI and Google are not our subprocessors for Henry: you choose them, and your key is your agreement with them. Their business API terms say they do not train on data sent through the API by default. [TEAM TO CONFIRM: wording on OpenAI and Google API terms.]4. No training on your content
We do not use your content, your questions or the AI’s answers to train AI models, and we do not let our providers do so.5. Who owns what
Between you and us, you own what you put in (your prompts, files and content) and what the AI produces for you. We own our own instructions, templates and the way the features work.6. Check what AI writes
AI makes mistakes. It can be wrong, out of date, or sound sure when it is not. You are responsible for checking AI output before you rely on it, publish it or share it. Oswald asks before he files anything into your Brand Files, and nothing he drafts goes live until it is published.7. What we keep
Your chats with Henry and Oswald are saved with your brand, so you can come back to them, and you can delete a conversation at any time. File descriptions are kept with the file. When a brand is erased, all of this is erased with it (see the Privacy Policy). Your own OpenAI or Gemini key is stored encrypted, and we only ever show its last four characters. You can change or remove it at any time.8. Connecting your own AI app
When you connect an AI app through the Atlas MCP, you sign in with your Brand Atlas login and choose Read or Read and write. From then on:- the app acts for you, with your permission, and what it reads goes to that app’s provider under your agreement with them;
- we are not responsible for what that app or its provider does with data it reads;
- everything it writes in Brand Atlas is done under your login, as if you had done it;
- you can see every connected app, and Revoke any of them, in your account settings under MCP. Revoking ends its access straight away. Log out of all devices also ends every connection.
9. Using AI fairly
Do not use the AI features to:- break the law, or anyone’s rights;
- make content the Acceptable Use Policy does not allow;
- get round limits, plans or safety measures, or try to make the AI reveal its instructions;
- send automated requests at a scale that harms the service for others.
Changes
- 4 October 2026: first draft.