What changed, and what did not.
Compliance calendar
The European dates that change what an AI call has to show.
Six of these rules apply now · one starts on 9 December 2026
Last reviewed: 6 October 2026
Seven European rules decide what an AI conversation has to disclose, record and account for. Six of them already apply: the DSA since 17 February 2024, MiCA since 30 December 2024, DORA since 17 January 2025, the Dutch Telecommunicatiewet since 1 July 2026, and both Article 50 and Article 57 of the AI Act since 2 August 2026. One is still ahead: the revised Product Liability Directive on 9 December 2026. A further date sits inside a rule that already applies — 2 December 2026, when the marking duty in Article 50(2) reaches systems that were already on the market before 2 August 2026.
The calendar
Seven rules, in the order their dates arrived.
Each row links to a page that sets out what the rule requires, with the article it comes from, and what Audact does about it. The two are kept apart on purpose: what the law demands carries a source, what we build carries a status.
- Date
- 17 February 2024
- Rule
- Digital Services Act, Regulation (EU) 2022/2065
- What it changes
- Transparency and notice-and-action duties for intermediary services.
- Status
- In force
- Date
- 30 December 2024
- Rule
- MiCA, Regulation (EU) 2023/1114
- What it changes
- Communications about crypto-assets must be fair, clear and not misleading.
- Status
- In force
- Date
- 17 January 2025
- Rule
- DORA, Regulation (EU) 2022/2554
- What it changes
- Operational resilience, and the ICT suppliers a financial entity depends on.
- Status
- In force
- Date
- 1 July 2026
- Rule
- Telecommunicatiewet (Netherlands)
- What it changes
- Consent for outbound calls to Dutch consumers has to be demonstrable.
- Status
- In force
- Date
- 2 August 2026
- Rule
- AI Act Article 50, Regulation (EU) 2024/1689
- What it changes
- A person must be told they are speaking to an AI system.
- Status
- In force
- Date
- 2 August 2026
- Rule
- AI Act Article 57
- What it changes
- Every Member State must have a national AI regulatory sandbox operational. The duty is on the Member State; for you it is an optional route.
- Status
- In force
- Date
- 9 December 2026
- Rule
- Product Liability Directive (EU) 2024/2853
- What it changes
- Putting your own name on someone else's product places you in the manufacturer's position.
- Status
- Ahead
Scope
Which of these actually reaches you.
Only one of the seven reaches everyone. Article 50 applies to any AI system that interacts with a person in the Union, whatever the sector and wherever the provider is based. The other six are conditional, and the condition is usually your client rather than you.
- DSA: only where the service is an intermediary service. A voice agent answering for one business is not one; a marketplace, a hosting service or a messaging platform is.
- DORA: only where your client is a financial entity listed in Article 2, or where you supply ICT services into one.
- MiCA: only where the conversation is about crypto-assets or a crypto-asset service.
- Telecommunicatiewet: only for outbound calls to people in the Netherlands.
- Product Liability Directive: from 9 December 2026, wherever you resell an AI service under your own name.
- Article 57 sandbox: a route you can choose, not a duty you carry.
Ahead
The two dates that have not arrived yet.
Article 50(2) only applies from 2 December 2026 to systems that were placed on the market before 2 August 2026. That transitional rule was inserted by the Digital Omnibus (EU) 2026/1744, in force 27 July 2026. It is a rule about existing systems: anything put on the market on or after 2 August 2026 is covered from day one.
On 9 December 2026 the revised Product Liability Directive applies. It is a directive, so the text you will be measured against is the national transposition, and that runs per Member State.
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
- Is it now illegal to run AI phone calls in the EU?
- No. Article 50 is a transparency duty: the person must be told, and synthetic output must be marked. Running an AI voice agent in the Union is lawful. What creates exposure is running one without the disclosure.
- My system went live before 2 August 2026. Does the December date help me?
- Only for the marking duty. Article 50(2) applies from 2 December 2026 to systems already on the market before 2 August 2026, under Article 111(4) inserted by the Digital Omnibus. The disclosure duty in Article 50(1) was not postponed.
- Which of these dates has already passed?
- Five dates, covering six of the seven rules: 17 February 2024 (DSA), 30 December 2024 (MiCA), 17 January 2025 (DORA), 1 July 2026 (Telecommunicatiewet) and 2 August 2026 (Article 50 and Article 57 of the AI Act, which share a date). All five are behind us, so all six rules apply now.
The disclosure belongs in the call, not in a policy document
Audact places the disclosure inside the conversation, at the position the policy for that country requires (Germany active · Netherlands in progress · Belgium, Austria queued), and records that it happened. See how the policy engine decides what is said, where, and when.
Related
Article 50, 2 August 2026 Product liability, 9 December 2026 Why white-label changes your position.
NL Telecomwet, 1 July 2026 Outbound calling into the Netherlands.
DORA, 17 January 2025 Financial entities and their ICT suppliers.
MiCA, 30 December 2024 What may be said about crypto-assets.
DSA, 17 February 2024 When an AI surface is an intermediary.
Article 57 sandboxes, 2 August 2026 Supervised testing. A route you can choose, not a duty you carry.
All frameworks The rules, one page each.
Check your own setup A few questions, one answer.
