AI & Your Data
If an AI step is part of your system, some of your information has to reach an AI service for it to be read. This page says plainly what goes where, what is kept and what you can ask me to change.
The Short Version
An AI step is sent the minimum it needs for the job, not your whole database.
Anthropic’s terms say they don’t train their models on inputs or outputs from their commercial products by default.
What went in, what came out and who approved it is logged inside your own system.
You are the controller of your data and I am your processor. We agree what each of us does before the build starts. I put that in a written data processing agreement with you, as UK GDPR requires.
The Details
I use Claude, made by Anthropic, through its commercial API. It is the only AI provider I use.
Your own system is hosted in the UK with 20i. When an AI step runs, the text it needs is sent to Anthropic. By default Anthropic may process requests in any of its available regions. It offers an option to keep processing on US infrastructure, and its data at rest is stored in the US. Anthropic’s agreement covers transfers of UK data using the UK Information Commissioner’s international data transfer addendum. See Anthropic’s data residency page and data processing addendum. I use Anthropic’s default settings.
Anthropic says it deletes API inputs and outputs from its systems within 30 days, with exceptions where the law requires it, where its Usage Policy needs enforcing, or where a separate agreement says otherwise. Inputs and outputs flagged as policy violations can be kept for up to two years. The details are on Anthropic’s retention page. In your own system, records and logs are kept while we work together, and deleted within 30 days of our work ending, or sooner if you ask.
No, not by default. Anthropic says it won’t use inputs or outputs from its commercial products to train its models unless you choose to allow it, with an exception for feedback you actively submit. I never submit your data to Anthropic as feedback. See Anthropic’s training page.
Yes. We decide together which fields are sent and which are left out, and that choice is written into the proposal. I won’t send payment card numbers, passwords or health information to an AI unless we’ve agreed it in writing.
Anthropic’s agreement says it must tell me about a security breach without undue delay, and within 48 hours of becoming aware of it. If it affects your data, I’ll tell you without undue delay, so you can meet your own duty to report it to the Information Commissioner’s Office within 72 hours. The privacy policy sets out how the Information Commissioner’s Office is told of a breach.
Ask before you commit to anything. See also AI agents and AI automation.