AI Automation and GDPR: What Your Business Must Arrange

You want an AI assistant to sort client emails, summarise an intake call or draft a payment reminder, which sends names, addresses and sometimes financial details through another company's software. The Dutch Data Protection Authority (Autoriteit Persoonsgegevens, AP) sees this go wrong: in August 2024 it reported a number of data breach notifications caused by employees entering patient or customer data into an AI chatbot.

For a judgement on your own situation, the AP or a privacy lawyer is the place to go.

Can you put customer data into an AI tool?

You can, under the same rules that apply to any other processing of personal data. The AP states that anyone who processes personal data with an algorithm must comply with the GDPR: with a legal basis, transparency towards customers, a purpose set in advance, as little data as possible and proper security. Which legal basis fits depends on the process, and using AI does not change it. Article 6 of the GDPR (Regulation (EU) 2016/679) lists, among others, performance of a contract, a legal obligation and legitimate interests; record for each processing activity which one you rely on.

You must also inform your customers. Article 13 GDPR requires you, when collecting data, to state among other things the purpose, the legal basis and the recipients, and whether you transfer data outside the EEA and whether an adequacy decision covers it; if the data does not come from the customer directly, Article 14 applies. When you add an AI service or a US provider, update that information, for example in your privacy notice.

For AI chatbots the AP separates two situations. When an employee enters personal data into a chatbot on their own initiative and against company agreements, that is a data breach. When the organisation allows the use, the AP does not call it a breach, and adds that it is often still not permitted by law. The AP advises setting out which data staff may enter, and arranging with the provider that input is not stored.

In an automated workflow you fix per step which fields go to an AI model, instead of leaving it to whatever someone pastes into a chat window.

When do you need a data processing agreement?

As soon as another party processes personal data on your behalf, Article 28(3) GDPR requires a contract that governs that processing. According to the AP, both parties are liable if it is missing. An AI workflow commonly involves several processors: the automation platform, your CRM, the AI model provider and your email provider. According to the AP, the agreement records at least:

• the subject, duration, nature and purpose of the processing, and which data about whom;
• that the processor acts only on your written instructions and does not use the data for its own purposes;
• confidentiality and security measures;
• that no sub-processor is added without your prior written permission;
• help with privacy requests, data breaches and a DPIA;
• deletion or return of the data afterwards, including copies.

With AI, the ban on own-purpose use raises whether a provider trains its model on your input, and the sub-processor rule raises which AI model your automation partner calls. More questions for an agency are in how to choose an AI automation agency.

What security does the GDPR require for an AI workflow?

Article 32 GDPR requires appropriate technical and organisational measures that match the risk, and names pseudonymisation and encryption among them. Article 5 adds that data must be "limited to what is necessary". In a workflow this comes down to four choices at the design stage:

• Only the fields each step needs. A step that sends an appointment confirmation needs a name and a time; the client file stays where it is.
• No personal data in logs or error messages. A log is a second copy of the data, with its own readers.
• Build and test with test data. Real customer data does not belong in a test environment.
• Limit access per connection. Each integration gets only the permissions its task requires.

In February 2026 the AP called on organisations not to use OpenClaw and similar AI agents on systems holding privacy-sensitive or confidential data, such as financial administration and employee data.

If a breach happens, Article 33 requires the controller to notify the supervisory authority "not later than 72 hours" after becoming aware of it where feasible, unless the breach is unlikely to result in a risk to the people concerned. Practical steps are in automation without compromising privacy.

When do you need a DPIA?

The AP states that a DPIA, a risk assessment carried out in advance, is mandatory when 2 or more of the 9 criteria on its DPIA list apply to your processing, including pilots, tests and trial projects. Separately, Article 30 requires a record of processing activities. Organisations employing fewer than 250 people are exempt, though the exemption is narrow: it lapses as soon as the processing is not occasional, is likely to result in a risk, or involves special categories of data or criminal records. A workflow that handles customer data every day is hard to call occasional.

Can you use US-based AI tools under the GDPR?

You can when the transfer rests on one of the routes in Chapter V of the GDPR (Article 44 onwards). For the United States the Data Privacy Framework is one such route, for participating organisations and the products it covers.

The AP's Dutch page on transfers outside the EEA, updated on 1 October 2026, still lists the United States among countries with an adequacy decision, limited to Data Privacy Framework participants. For those transfers the AP describes two situations:

• The organisation participates. You may transfer the data without another transfer tool and without additional measures. Participation does not automatically mean every product of that organisation is covered, so the AP advises checking per product.
• The organisation does not participate. You then need another tool, such as standard contractual clauses (the clauses adopted by the Commission under Article 46(2)(c)), and where needed additional measures such as encryption or pseudonymisation.

An example from our own setup: HighLevel, the company behind our CRM GoHighLevel, states in its privacy policy that it participates in the EU-U.S. Data Privacy Framework. By the AP's standard, the per-product question stays open.

Why is the Data Privacy Framework under scrutiny?

The framework, which rests on Commission Implementing Decision (EU) 2023/1795, is under pressure from a pending appeal at the EU Court of Justice and a US ruling on the independence of the regulator it relies on.

1. An appeal at the Court of Justice. On 31 October 2025 Philippe Latombe lodged an appeal against a General Court judgment of 3 September 2025, asking the Court to annul the decision after all (Case C-703/25 P).
2. A ruling in the US. On 29 June 2026 the US Supreme Court held in Trump v. Slaughter that the President may remove members of the Federal Trade Commission. The 2023 decision explicitly refers to the independence of that regulator. On 31 July 2026 the European Data Protection Board therefore asked the Commission in a letter "to closely assess whether this development affects the functioning" of the decision.

As long as the decision stands, transfers to participating organisations remain allowed; if it falls, you need another Chapter V route, for example standard contractual clauses with additional measures.

What does the EU AI Act change about the GDPR?

The GDPR remains the framework. Article 2(7) of Regulation (EU) 2024/1689 states that the AI Act does not affect it. That sentence carries a reservation: originally for Article 10(5) and Article 59, and since the amendment by Regulation (EU) 2026/1744, known as the Digital Omnibus, "Without prejudice to Articles 4a and 59". Articles 4a and 59 allow additional processing as an exception and under strict conditions: of special categories of personal data to detect and correct bias in AI (Article 4a), and within an AI regulatory sandbox (Article 59). Who may do so and under which conditions is set out in the amended text on the European Commission's AI Act Service Desk.

How does Mindsora handle this for clients?

At Mindsora we sign a data processing agreement with every client before implementation; Article 13 of our terms and conditions (in Dutch) commits both parties to one whenever we process personal data for a client. We build our automations in n8n, on a Hostinger server in Frankfurt. As our CRM we use GoHighLevel; to the extent HighLevel transfers data to the US or makes it accessible there, the US route above applies to that link. Our privacy statement (in Dutch) lists the processors we share personal data with and their locations.

More in how we automate at Mindsora and on our services page.

To find out which processes in your business lend themselves to automation and which data would be involved, take the free AI Readiness Scan.