Milōn uses an AI model (Claude, from Anthropic). Identifiers are stripped before model calls; amounts stay. This page records what that means.
Effective 4 September 2026
When you ask the in-app assistant a question, or when the platform needs model help to talk about a business, identifiers are stripped from what leaves Milōn for Claude. Amounts stay, so workings and accountant sign-off use real figures.
What remains is enough for the assistant to be useful — industry, the shape of the ratios, and the actual figures needed for workings — without the company name, tax IDs, or account numbers that would identify the business.
The model is Claude, supplied by Anthropic. That is a processing fact, not an endorsement line. The same statement appears in the site footer.
Where AI drafts an advisory note for an accountant, a human reads it and signs it before a client ever sees it. The model does not send mail to your clients. It does not file anything with SARSthe IRS or a state department of revenue. It does not replace the accountant you already have.
The score, the waterfall, the forecast, and the action plan do not require you to open the assistant. Asking a question is optional. Uploading figures into your own workspace is how the product gets its arithmetic — that store is covered by the privacy notice, under row-level security, visible only to you and the people you invite.
Privacy covers what we store. Terms cover how the workspace may be used.