When did Article 50 start applying?

2 August 2026. The Digital Omnibus on AI deferred the high-risk obligations to December 2027 and August 2028 but left Article 50 on its original timeline; only the paragraph 2 marking obligation for providers was postponed, to 2 December 2026. Content created before August does not have to be labelled retroactively, though the Commission encourages it.

What does 'unless it is obvious' mean in practice?

It is read narrowly. The test is whether a reasonably well-informed, observant and circumspect person would realise they are dealing with an AI. On an inbound phone call answered in a natural human voice, most callers would not, so in practice you should assume disclosure is required.

Does the disclosure have to be at the start of the call?

The Commission's guidance is that users are informed that they are interacting with an AI system from the start of the interaction, and that the disclosure is clear, distinguishable and accessible. A line in a terms document a caller never sees would not meet that.

Are there situations where no disclosure is needed?

The Commission lists machine-to-machine communication, systems operating purely in the background, cases where the AI interaction is genuinely obvious, and standard editing assistance. A customer-facing phone agent is none of those.

Article 50 and your AI phone agent

Article 50 of the EU AI Act requires that people be told they are dealing with an AI system, unless that is obvious to a reasonably well-informed person. It has applied since 2 August 2026. For an AI phone agent the duty sits mainly with the provider of the system, not the business deploying it.

General guidance, not legal advice. This page summarises publicly available material on Regulation (EU) 2024/1689 and links the primary sources. It is not a substitute for advice from a qualified lawyer on your own circumstances. Confirm anything you intend to rely on with counsel.

Last reviewed 25 August 2026.

What the Article actually says

Article 50 of Regulation (EU) 2024/1689 — the AI Act — is the transparency article. It does not ban anything. It requires that certain AI systems make themselves known. The Article text is published here, and the European Commission maintains an official FAQ on it.

The paragraph that matters for a phone agent is the first. Providers of AI systems intended to interact directly with people must ensure those people are informed that they are interacting with an AI system — unless that is obvious to a natural person who is, in the Article's words, reasonably well-informed, observant and circumspect.

The remaining paragraphs cover synthetic content marking, emotion recognition and biometric categorisation, and deepfakes and AI-generated text. Most of those do not arise for an agent whose job is to answer a call and book an appointment.

The obligations are split by role, and the split surprises people

Paragraphs 1 and 2 bind providers — the party that develops the system, or has it developed, and puts it on the market under its own name. Paragraphs 3 and 4 bind deployers — the business using the system under its own authority in a professional capacity.

So if you buy an AI receptionist rather than build one, the "tell them it is an AI" duty is not primarily yours. It belongs to your vendor. That is the opposite of what a lot of the marketing around the deadline implies.

It does not make you indifferent to it. If your vendor fails to disclose, it is still your phone line, your customers and your reputation, and other law — data protection, national consumer rules — applies to you regardless. We have set out the split in more detail, including the case where customising an agent heavily can pull you into the provider role.

The 2 December 2026 date, and what it is not

Article 50 has applied since 2 August 2026. The Digital Omnibus on AI, in force since 27 July 2026, deferred the high-risk obligations — to 2 December 2027 for stand-alone Annex III systems and 2 August 2028 for AI embedded in regulated products — but left Article 50 on its original timeline. The one exception is paragraph 2: the marking obligation for providers was postponed to 2 December 2026.

It is not a general extension of the disclosure duty. If you are putting an AI phone agent on your line now, the interaction-disclosure obligation is live today, not in December.

Whether the machine-readable marking obligation reaches the audio a voice agent produces during a live call is, in our reading, genuinely unsettled — the marking rules are written with generated media in mind rather than real-time conversation. We are flagging that rather than answering it. If your use case turns on that question, it is one for your own counsel.

What good practice looks like on a phone line

The Commission's guidance is that disclosure should be understandable and perceivable without special tools, clear and distinguishable, and given from the start of the interaction. On a voice channel that means saying it, in the greeting, in the language of the call.

It does not require a legal recital, and it does not require the agent to sound robotic. A single natural clause in the opening line satisfies the substance. We have written out openings that work in English, German, French and Dutch, alongside the common wordings that fall short.

Two practical habits are worth adopting regardless of where the legal line falls: keep a recording or transcript showing the disclosure was made, and make sure a caller who asks for a human gets one.

Common questions

Does Article 50 mean my AI receptionist must say it is an AI?: Article 50 of the EU AI Act requires that people be told they are dealing with an AI system, unless that is obvious to a reasonably well-informed person. It has applied since 2 August 2026. For an AI phone agent the duty sits mainly with the provider of the system, not the business deploying it.
When did Article 50 start applying?: 2 August 2026. The Digital Omnibus on AI deferred the high-risk obligations to December 2027 and August 2028 but left Article 50 on its original timeline; only the paragraph 2 marking obligation for providers was postponed, to 2 December 2026. Content created before August does not have to be labelled retroactively, though the Commission encourages it.
What does 'unless it is obvious' mean in practice?: It is read narrowly. The test is whether a reasonably well-informed, observant and circumspect person would realise they are dealing with an AI. On an inbound phone call answered in a natural human voice, most callers would not, so in practice you should assume disclosure is required.
Does the disclosure have to be at the start of the call?: The Commission's guidance is that users are informed that they are interacting with an AI system from the start of the interaction, and that the disclosure is clear, distinguishable and accessible. A line in a terms document a caller never sees would not meet that.
Are there situations where no disclosure is needed?: The Commission lists machine-to-machine communication, systems operating purely in the background, cases where the AI interaction is genuinely obvious, and standard editing assistance. A customer-facing phone agent is none of those.

Provider or deployer — which are you?
The obligations split by role, and most businesses are not the party the disclosure duty binds.
What a compliant call opening sounds like
Scripted openings in English, German, French and Dutch, and the ones that fall short.
AI voice agents for EU businesses
How AI Placers works across the 24 EU languages, and how we handle disclosure.

Want to hear a compliant opening?

We will run our agent against your own call scenarios during a free two-week proof-of-concept, disclosure included, so you can judge it on your line rather than in a demo.

Book a proof-of-concept