In 30 seconds

Run this before an AI workflow goes near personal or confidential data. It is deliberately short. If you cannot tick an item, that is the work to do before launch, not after. Tick each only when you have a real answer, not an assumption.

What goes in

  • The agent receives only the data the task actually needs.
  • Secrets, passwords, and payment details are excluded entirely.
  • Special-category data (health, ethnicity, and similar) is excluded unless specifically cleared.
  • Where the real identifier is not needed, it is masked or replaced with an ID.

Where it goes

  • You know which vendor and which model provider the data reaches.
  • The vendor does not train on your data, confirmed in writing.
  • You know where the data is stored and for how long.
  • You can have the data deleted on request.

Who sees it

  • The agent's outputs are stored somewhere with controlled access.
  • Only the people who need to can read the results.
  • Any logs that capture the data are covered by the same rules.

Promises and rules

  • Using AI here does not break a promise you made to customers or staff.
  • The workflow is consistent with your privacy notice.
  • For high-volume or high-sensitivity data, you have taken specific advice.

When it goes wrong

  • You know what happens if the agent mishandles data, and who is told.
  • There is a way to switch the workflow off quickly if needed.

What this does not solve

A ticked checklist reduces risk; it is not legal sign-off. Your obligations depend on where you operate and the data involved. Use this to catch the everyday gaps, and get specific advice for the hard cases. The detail behind each item is in how to keep customer data safe with AI.

Good enough to move on when

  • Every item is ticked with a real answer, not a hope.
  • The few you could not tick have owners and dates.
  • The answers are written down next to the workflow.