AI Automation and GDPR: A 10-Point Checklist | Loop2Tech Skip to main content

New on the blog: Google Maps Ranking Factors in 2026: What Moves a Business Into the Top 3 Read it

AI Automation and GDPR: A 10-Point Privacy Checklist Before You Build

A golden padlock resting on a computer keyboard.

Quick answer: Before an AI automation touches customer data, sign a data processing agreement with whoever builds and runs it, name every AI provider in it, check where the data goes, keep personal data out of prompts where you can, label chatbots as AI, and limit what an AI step can send or spend. This is practical guidance from a studio that builds these workflows, not legal advice; have a lawyer check your contracts.

Automation projects usually start with "can it reply to our customers?" and only later ask "where does that customer's message go?". By then the workflow is live and fixing it is harder.

My opinion: treat every AI automation as a data project first and a time-saver second. The checks below take an afternoon, and they are much cheaper than explaining a leak to your customers.

The checklist

  1. Sign a data processing agreement. Under GDPR Article 28, a provider handling your customers' data needs a written contract, and needs your authorisation before using sub-processors such as AI vendors.
  2. Name the AI providers. OpenAI, Anthropic or Google become sub-processors. List them.
  3. Check international transfers. If anyone outside the EU or UK can access the data, the transfer needs safeguards. Pakistan, for example, is not on the EU adequacy list, so EU data accessed from there needs standard contractual clauses or similar.
  4. Know the AI vendor's terms. OpenAI and Anthropic say they do not train on business API data by default, and both keep API data for up to about 30 days. A court order can override that, as one did for OpenAI in 2025.
  5. Minimise what you send. Strip names, emails and payment details from AI prompts where the task does not need them.
  6. Label chatbots as AI. The EU AI Act's Article 50 has required this since 2 August 2026.
  7. Guard against prompt injection. The OWASP Top 10 for LLM applications ranks it first. An AI step that reads customer messages should not also be able to send your data anywhere it likes.
  8. Cap spending and actions. Set limits on API spend and on what an AI step may do on its own.
  9. Keep secrets safe. API keys belong in the platform's credential store, never in a shared document or workflow note.
  10. Plan for change. AI models are retired on notice, and US states including Colorado and California have new automated decision rules starting 1 January 2027. Review each workflow at least every quarter.

If you are deciding what to automate, start with AI automation for small businesses. For support bots specifically, see AI customer support on Shopify.

Where I could be wrong

  • Rules are moving. The EU amended parts of the AI Act in 2026 and US states are still writing theirs. Check current guidance for your country.
  • Small, low-risk workflows need less. Copying order numbers between two of your own tools is not the same risk as an AI reading customer emails.
  • I am not a lawyer. This list reflects what we do on our own projects. A lawyer should review your data processing agreement and liability terms.

How Loop2Tech helps

Our Automation Audit, US$199, includes a data and privacy check of every tool a workflow touches, and we name each AI provider in the data processing agreement before we build. Automation Care from US$150 a month includes a quarterly review. See every AI automation package, our AI automation service, or ask us about your data.

Frequently asked questions

Is AI automation GDPR compliant?

It can be, with a data processing agreement, named sub-processors, safeguards for international transfers, minimal personal data in prompts and clear records of what each workflow does.

Do AI providers train on my customers' data?

OpenAI and Anthropic say they do not train on business API data by default, and keep it for up to about 30 days. Check the current terms of whichever provider you use.

Does my chatbot have to say it is AI?

In the EU, yes. Article 50 of the EU AI Act has required chatbots to disclose that they are AI since 2 August 2026.

What is prompt injection?

A message or document that tricks an AI step into ignoring its instructions, for example to reveal or send data. OWASP ranks it the top risk for AI applications.

Do I need a lawyer for an AI automation project?

For anything touching customer personal data, have a lawyer review the data processing agreement, transfer clauses and liability terms. This checklist is not legal advice.