What changed on the day it started.
NL Telecomwet
Calling Dutch consumers with an AI agent: what has to be true first.
Last reviewed: 6 October 2026
The Dutch Telecommunicatiewet decides whether you may place the call at all. Article 11.7 permits an automated calling or communication system without human intervention only where the subscriber has given prior consent, and it puts the burden of showing that on the sender. An AI voice agent is exactly such a system. This is a separate question from Article 50 of the AI Act, which decides what has to be said once the line is open.
What the law requires
- An automated calling or communication system used without human intervention may only carry unsolicited commercial communication where the subscriber has given prior consent; the sender must be able to demonstrate that consent.
- Telecommunicatiewet, art. 11.7(1)
- Consent has to be specific and informed, and the party relying on it carries the burden of proof.
- Telecommunicatiewet, art. 11.7(1) · Regulation (EU) 2016/679, Art. 7(1)
- Every communication must identify on whose behalf it is sent and offer a means, free of charge, to withdraw consent or object to further contact.
- Telecommunicatiewet, art. 11.7
- The amended Telecommunicatiewet has applied since 1 July 2026, one month before the AI Act transparency duties started to apply.
- Telecommunicatiewet, as amended, in force 1 July 2026
- The ACM supervises and can impose an administrative fine of up to €900,000 per violation, or 10% of annual turnover where that is higher.
- Telecommunicatiewet, art. 15.4
- This sits alongside, not instead of, the AI Act. Consent decides whether the call may be placed; Article 50 decides what must be said on it.
- Regulation (EU) 2024/1689, Art. 50(1)
What Audact does about it
- Consent is checked before dialling. Without a recorded prior-consent flag the call is not placed: the gate sits in front of the dialler, not in a report afterwards.
- In controlled testing · Germany: national policy layer active. Netherlands: in development. Belgium and Austria: planned.
- Time-window rules are enforced at the proxy layer, so a mistake in a campaign tool cannot dial through them.
- In controlled testing · Germany: national policy layer active. Netherlands: in development. Belgium and Austria: planned.
- Consent is captured and held by you; the receipt does not establish that it was validly obtained. Before dialling, the platform checks the calling hours for the call's jurisdiction today.
- In development
- The Dutch policy pack (the national layer on top of Article 50) is being built to the same verification standard as the German one.
- Germany: active · Netherlands in progress
Two questions
May you call, and what must you say once connected?
These are separate duties from separate statutes, and satisfying one says nothing about the other. The Telecommunicatiewet governs the act of contacting someone: prior consent, identification, an easy way out. Article 50 of the AI Act governs the content of the interaction: the person is told they are speaking to an AI.
A campaign can be perfectly disclosed and still unlawful because no consent was recorded, and perfectly consented and still exposed because the agent never said what it was. Both gates have to be closed, and they close at different moments: one before the number is dialled, one in the first seconds of the conversation.
Demonstrability
The burden of proof is the operative word in article 11.7.
The article does not only require consent; it requires that the sender can show it. In practice that means four things per call: who consented, when, for what, and through which channel. A consent that exists in a CRM but cannot be tied to the call that was placed is a consent you cannot rely on when the ACM asks.
The receipt for a call carries the policy evaluation that preceded it; consent itself is not checked by the platform today.
Note
Information, not legal advice.
This page sets out the rules as we read them, with the article numbers so you can check us. It is not legal advice, and your own position depends on facts this page cannot know. Take advice from qualified counsel before you rely on it.
Questions
- Does an existing customer relationship let me call without fresh consent?
- Do not assume it does. Article 11.7 is written around prior consent and puts the burden of proof on the sender, and the exceptions are narrow and drawn per channel. Establish which basis you are relying on per campaign, and record it, rather than relying on a general customer relationship.
- Does the AI have to say it is an AI on a Dutch call?
- Yes, but that duty comes from Article 50 of the EU AI Act, not from the Telecommunicatiewet. Both apply to the same call and neither one covers the other.
- What can the ACM fine?
- Up to €900,000 per violation, or 10% of annual turnover where that is higher, under article 15.4 of the Telecommunicatiewet. Fines can be imposed per campaign rather than once per company.
- We are not established in the Netherlands. Does it still apply?
- The rule follows the person being called, not the company placing the call. If the subscriber is in the Netherlands, the Dutch rules are the ones the ACM will read your campaign against.
The consent gate sits in front of the dialler
Voice shows where the check happens: before the number is dialled, inside the flow, and what the policy evaluation looks like in the record.
Related
The 1 July 2026 date EU AI Act Article 50 What has to be said once connected.
ePrivacy The European rule this national layer sits on.
GDPR Consent as a lawful basis, and its proof.
Check your own setup A few questions, one answer.
All frameworks
