AI and your data
AI in Studio reads, sorts and drafts so your team does not have to. It only ever sees what you approve, with personal details hidden first.
Last updated 26 October 2026
01Three settings, chosen per account
| Setting | What happens |
|---|---|
| Claude, details hidden | Personal details are hidden in your browser, you check the request, then it goes to Claude by Anthropic. The real details are put back into the answer in your browser. |
| On this device only | A smaller AI model downloads once and runs inside your browser. Nothing you ask is sent anywhere. It is slower and less capable. |
| Off | AI is hidden. Every tool still works on its own. |
02What is hidden before anything is sent
Before a request leaves your browser, Studio swaps these for placeholders such as [PERSON_1]:
- names of people and companies
- email addresses, phone numbers, addresses and postcodes
- dates of birth, passport, driving licence and NHS numbers
- National Insurance numbers and UTRs
- sort codes, account numbers, IBANs and card numbers
- anything typed into a name, client or contact field
Files that look like they contain health, criminal-record or other special category information are held back from Claude, and a person is told. The link between placeholders and real details stays in your browser. Details are found by recognising patterns, which catches most but cannot promise to catch everything, so the check below matters.
03You check every request
Studio shows you exactly what will be sent and waits for you to confirm. You can hide anything else it missed before sending, or cancel and nothing is sent. You can choose to skip the check for the rest of a browser tab.
04AI never does the maths
When AI reads a payslip or fills in a form, the figures always come from the tool's own tested calculation, never from the AI. AI can still misread a document, so check anything you rely on.
05What Anthropic does with requests
Anthropic provides Claude to Mondas under its commercial terms and data processing agreement, which do not allow requests to be used to train models. Requests are processed outside the UK. Mondas does not store or log their content.
06AI employees and workflows
- An AI employee works on a task you start, one step at a time, with the same hiding of personal details. One check before sending covers each task.
- The files you give an employee and its history stay in your browser, on your device. Employees only work while their page is open.
- With Mondas Agent, employees work on your own computer and files stay there. Linked to your account, they think with Claude, with the same hiding of personal details; you agree to this once when you set it up. Or choose an AI model that runs on the computer, offline, so nothing is sent.
- Email is optional. If you connect an email account to Mondas Agent, an employee only takes work from the people you list, and only when their email provider confirms it is really from them. It replies only to them.
- Workflows use the same hiding and the same check before anything is sent.
07Daily limits
During a free trial, Claude can be used up to 20 times per tool each day, and 150 times in total. Each file in a batch counts once. If a request would go over, Studio says so before anything is sent, with how many are left. Paying accounts have no daily limits.
