Since 2 August 2026, Article 50 of the EU AI Act has been the rule that decides how an AI sales or support agent must introduce itself to anyone in the EU. Most of the noise around the Act concerns high-risk systems — but the Digital Omnibus (Regulation (EU) 2026/1744) deferred those obligations, so for teams running AI agents that talk to EU prospects and customers, the live compliance question today is transparency, not high-risk conformity. This checklist covers what Article 50 requires right now, what it does not, and what a sales or support team should actually do about it. It is general information, not legal advice — get counsel for your specific setup.
Quick answer: Article 50 of the EU AI Act applies from 2 August 2026 and requires that people be told they are interacting with an AI system unless that is obvious, that AI-generated content carries machine-readable marking, and that deepfakes — including cloned voices presented as real people — are disclosed. It reaches non-EU companies whose AI output is used in the EU. High-risk (Annex III) obligations were deferred to 2 December 2027 by the Digital Omnibus, so for most AI sales and support agents the current obligation is disclosure and marking, backed by significant administrative fines for non-compliance.
Where Zian stands
Zian’s agents operate across phone, SMS, email and WhatsApp in 30+ languages, and voice cloning is supported — which is exactly why we treat disclosure as a platform behaviour, not a customer afterthought. Honest disclosure is also the commercially smart move: a prospect who discovers mid-call that “Alex” was a bot is not warming to your demo. If you want AI sales agents built with transparency as a default, Apply For Partnership.
Why this is a transparency question, not a high-risk one (for now)
The EU AI Act (Regulation (EU) 2024/1689) layers obligations by risk. Prohibited practices have applied since 2 February 2025, and the Digital Omnibus added two more prohibitions, on AI systems that generate or manipulate realistic intimate or sexually explicit material of an identifiable person without that person’s explicit consent, or child sexual abuse material (Article 5(1)(ba) and (bb), the latter subject to a national-law “without right” defence). New Article 5(1a) confines both: placing such a system on the market is prohibited only where that output is its intended purpose, or is a reasonably foreseeable outcome the system lacks adequate safeguards to prevent, and using one only where the deployer uses it for that purpose. They apply from 2 December 2026. The general application date — including Chapter IV, where Article 50 sits — arrived on 2 August 2026. The heavyweight regime, for high-risk systems listed in Annex III, was originally due the same day.
Then the Digital Omnibus intervened. The Commission proposed it on 19 November 2025, a political agreement followed on 7 May 2026, and Regulation (EU) 2026/1744, signed on 8 July 2026 and published in the Official Journal on 24 July 2026, entered into force on 27 July 2026. It pushed the application of Annex III high-risk obligations to 2 December 2027, and obligations for high-risk AI embedded in Annex I regulated products to 2 August 2028. We covered the deferral in detail in our EU AI Act August 2026 status post — this article is the practical companion: the checklist for the part that did not move.
Article 50 was not deferred. It applies now, with one narrow transition: the Omnibus inserted a new Article 111(4), under which providers of AI systems “generating synthetic audio, image, video or text content, that have been placed on the market before 2 August 2026 shall take the necessary steps in order to comply with Article 50(2) by 2 December 2026”. The European Commission’s Article 50 transparency FAQ confirms the grace period covers only that marking duty, and adds that content generated before 2 August 2026 “does not need to be labelled retroactively”. The Omnibus made one other change to Article 50 itself: it replaced paragraph 7, on codes of practice (see the status section below). As at 12 October 2026 we found no further amendment to Article 50 on EUR-Lex or the Commission’s AI Act pages.
Article 50 in full: all seven paragraphs and their exceptions
Article 50 has seven paragraphs, and four of them carry exceptions that summaries tend to drop. This is the instrument as amended by Regulation (EU) 2026/1744, read on the Publications Office text of Regulation (EU) 2024/1689 and the Omnibus on 12 October 2026.
| Paragraph | Who it binds | The duty | Exceptions and limits in the text |
|---|---|---|---|
| 50(1) | Providers of AI systems “intended to interact directly with natural persons” | Design the system so the people concerned “are informed that they are interacting with an AI system” | Not needed where it is “obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use”; does not apply to systems authorised by law to detect, prevent, investigate or prosecute criminal offences, subject to appropriate safeguards for the rights and freedoms of third parties, unless they are available for the public to report a criminal offence |
| 50(2) | Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text | Outputs “marked in a machine-readable format and detectable as artificially generated or manipulated”, with technical solutions that are effective, interoperable, robust and reliable as far as technically feasible | Does not apply to an assistive function for standard editing, to systems that do not substantially alter the deployer’s input or its semantics, or where authorised by law for criminal-offence purposes; systems on the market before 2 August 2026 have until 2 December 2026 (Article 111(4)) |
| 50(3) | Deployers of an emotion recognition or biometric categorisation system | Inform the people exposed of the system’s operation, and process the personal data under the GDPR, Regulation (EU) 2018/1725 and Directive (EU) 2016/680, as applicable | Does not apply to such systems permitted by law to detect, prevent or investigate criminal offences, subject to safeguards and Union law |
| 50(4), first limb | Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake | Disclose that the content has been artificially generated or manipulated | Does not apply where the use is authorised by law for criminal-offence purposes; for evidently artistic, creative, satirical, fictional or analogous work, limited to disclosing the content’s existence in a way that does not hamper the work |
| 50(4), second limb | Deployers of an AI system that generates or manipulates text published to inform the public on matters of public interest | Disclose that the text has been artificially generated or manipulated | Does not apply where authorised by law for criminal-offence purposes, or where the text has undergone human review or editorial control and a natural or legal person holds editorial responsibility for the publication |
| 50(5) | Everyone under paragraphs 1 to 4 | Information given “in a clear and distinguishable manner at the latest at the time of the first interaction or exposure”, conforming to accessibility requirements | None |
| 50(6) | Everyone under paragraphs 1 to 4 | Paragraphs 1 to 4 do not affect the Chapter III high-risk requirements, and are without prejudice to other transparency obligations laid down in Union or national law for deployers of AI systems | Not an exception: other disclosure laws still apply on top |
| 50(7), as replaced by the Omnibus | The Commission | Encourage codes of practice on “detection, marking and labelling” of generated content and assess, taking utmost account of the AI Board’s opinion, whether adhering to them is adequate for paragraphs 2 and 4 | If a code is inadequate, the Commission may adopt an implementing act with common rules |
Article 50(4) in full, because it is the limb that catches cloned voices and is the one most often paraphrased too broadly:
“4. Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated. This obligation shall not apply where the use is authorised by law to detect, prevent, investigate or prosecute criminal offence. Where the content forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme, the transparency obligations set out in this paragraph are limited to disclosure of the existence of such generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work. Deployers of an AI system that generates or manipulates text which is published with the purpose of informing the public on matters of public interest shall disclose that the text has been artificially generated or manipulated. This obligation shall not apply where the use is authorised by law to detect, prevent, investigate or prosecute criminal offences or where the AI-generated content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content.”
Article 3(60) defines a deep fake as “AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful”. The definition is not limited to people: a synthetic recording of a real place or event can qualify too.
Article 50 obligations at a glance
| Obligation | Who it applies to | Applies from | What a sales/support team actually does |
|---|---|---|---|
| Art 50(1): tell people they are interacting with an AI system, unless obvious | Providers of AI systems that interact directly with natural persons | 2 Aug 2026 | Put the disclosure in the agent’s opening line, chat header and email footer; don’t rely on “obvious” |
| Art 50(2): mark synthetic audio, image, video and text as artificially generated, in machine-readable form | Providers of generative AI systems | 2 Aug 2026 (systems on market before that date: 2 Dec 2026) | Ask your vendor in writing how outputs are marked/watermarked and keep the answer |
| Art 50(3): inform people exposed to emotion recognition or biometric categorisation, and process their data under the GDPR | Deployers of those systems | 2 Aug 2026 | If you score caller emotion or categorise people biometrically, notify them — and first check the Article 5 prohibitions |
| Art 50(4): disclose deep fakes (image, audio or video) and AI-generated public-interest text | Deployers | 2 Aug 2026 | Label cloned-voice audio that resembles a real person; label AI-written public-interest text unless it has had human review or editorial control and someone holds editorial responsibility |
| Annex III high-risk obligations (risk management, logging, conformity assessment) | Providers/deployers of listed high-risk systems | Deferred to 2 Dec 2027 | Diarise it; assess whether your use case is even Annex III (most sales/support agents are not) |
The checklist
1. Disclose the AI — Article 50(1)
Article 50(1) requires providers to ensure that AI systems intended to interact directly with natural persons are designed so those people are informed they are dealing with an AI system — “unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use”, as Article 50(1) puts it. The Commission’s Article 50 Q&A says the exception “should be interpreted in a restrictive manner, given that it deprives people of transparency”. Our reading: on a phone call, nothing about the medium makes AI obvious, so a voice agent that sounds human needs to say so.
Do this now:
- Script it into the greeting. The disclosure belongs at the start of the first interaction, not three turns in. Our disclosure scripts post has wording patterns that announce the AI without killing the conversation.
- Cover every channel. Voice, chat widget, SMS, email, WhatsApp — each needs its own disclosure surface. A chat window labelled “AI assistant” in the header may do the job; a voice call needs the words said out loud.
- Don’t hide it in terms of service. Article 50(5) requires the information to be provided in a clear and distinguishable manner at the latest at the time of the first interaction, and to meet accessibility requirements.
- Bake it into guardrails. Treat “always disclose, never deny being an AI when asked” as a hard rule in the agent’s instructions — see our guide to writing guardrails for autonomous agents.
2. Synthetic content marking and cloned voices — Articles 50(2) and 50(4)
Article 50(2) sits with the provider of the generative system: outputs must be “marked in a machine-readable format and detectable as artificially generated or manipulated”, per the Commission FAQ. If you buy rather than build, your main job is vendor diligence: get a written statement of how the platform marks generated audio and text, and whether it adheres to the Code of Practice discussed below.
Article 50(4) sits with you, the deployer. It covers deep fakes — under Article 3(60), AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. A cloned voice of your founder fronting outbound calls is squarely in this territory: the person hearing it should be told the voice is AI-generated. Consent from the person cloned is a separate, additional issue — see our voice cloning consent and compliance guide. Article 50(4) also requires disclosure of AI-generated text published to inform the public on matters of public interest, unless it has undergone human review or editorial control and a natural or legal person holds editorial responsibility for the publication. Both limbs fall away where the use is authorised by law to detect, prevent, investigate or prosecute criminal offences, and for evidently artistic, creative, satirical or fictional work the duty shrinks to disclosing that generated content exists. The full text is quoted above.
3. Timing and wording patterns
Article 50(5) fixes the timing rule: clear, distinguishable, and at the latest at the first interaction or exposure. Patterns that work in practice:
- Voice: “Hi, this is Alex, an AI assistant calling on behalf of [company]” — identity, AI status and principal in one breath, before the pitch.
- Chat: persistent label in the widget header plus a first-message statement; don’t rely on a favicon.
- Email/SMS: a plain-language line identifying the sender as an AI assistant of the named company.
- Handoffs: when a human takes over, say so — the boundary between AI and human should stay legible in both directions.
If you already run disclosure scripts for US state laws or Australian expectations, extend them rather than forking: one honest global greeting is easier to audit than per-jurisdiction variants, and jurisdictions are converging — see our breakdown of Colorado’s SB 26-189 for the US side.
4. What records to keep
Article 50 does not impose the formal logging regime that high-risk systems will face — but market surveillance authorities can ask you to show compliance, and Article 99 backs the transparency obligations with significant administrative fines (Article 99(4)(g) lists Article 50 breaches expressly; we deliberately don’t quote amounts). Sensible evidence to keep:
- Versioned disclosure scripts and screenshots of chat/email disclosure surfaces, with deployment dates.
- Call recordings or transcripts sampled to show the disclosure is actually delivered (subject to your recording-consent obligations — see our outreach compliance overview).
- Vendor statements on Article 50(2) marking, and any Code of Practice adherence.
- Guardrail and approval-gate configurations showing disclosure is enforced, not optional — human-in-the-loop controls help here.
- A dated note of your Annex III self-assessment, so the 2 December 2027 deadline doesn’t surprise you.
Guidelines and Code of Practice: current status
Two support instruments are now final — neither is a draft anymore. On 20 July 2026 the European Commission published the Guidelines on the Article 50 transparency obligations, which it says “clarify which providers and deployers must comply with the transparency obligations for interactive AI systems and the marking and labelling of AI-generated content”; its Q&A adds that they give “further guidance and practical examples” on the “obvious” exception. And the Code of Practice on Transparency of AI-generated Content — final version published 10 June 2026 — has been confirmed by the Commission and the AI Board as an adequate voluntary tool for demonstrating compliance, with “about 190” signatories by the end of July 2026, per the Commission’s code page (last updated 31 July 2026 when we read it on 12 October 2026). Under Article 50(7) as replaced by the Omnibus, the Commission no longer approves such codes by implementing act; it assesses whether adherence is adequate for paragraphs 2 and 4. Signing is optional; if your vendor has signed, that is useful evidence for your records, though not conclusive proof of compliance.
Does this reach a US or Australian company?
Yes, in most realistic setups. Article 2(1) applies the Act to providers placing AI systems on the market in the Union irrespective of where they are established, and — the clause that catches outbound teams — to “providers and deployers of AI systems that have their place of establishment or are located in a third country, where the output produced by the AI system is used in the Union” (Article 2(1)(c)). An Australian company whose AI agent calls prospects in Berlin is producing output used in the Union. The practical read: if EU numbers or EU inboxes are in your target list, run the checklist.
FAQs
Does the EU AI Act ban AI sales calls?
No. The AI Act does not prohibit AI sales or support calls. Its Article 5 prohibitions target practices like subliminal manipulation and exploiting vulnerabilities, not ordinary commercial outreach. For AI agents talking to EU users, the operative duty today is Article 50 transparency: disclose the AI, mark synthetic content, and label deepfakes. Other rules, such as GDPR and national telemarketing laws, still apply separately.
Do I have to disclose the AI if it’s obvious?
Article 50(1) contains an exception where the AI interaction is obvious to a natural person who is reasonably well-informed, observant and circumspect, and the European Commission’s guidance says this exception is interpreted restrictively, per the Commission’s Article 50 transparency FAQ. A human-sounding voice agent on a phone call is not obvious, so disclose. In practice the safest and cheapest policy is to disclose everywhere and skip the argument.
Does Article 50 apply to a US or Australian company calling EU prospects?
Generally yes. Under Article 2(1) the AI Act applies to providers placing AI systems on the EU market irrespective of where they are established, and to providers and deployers located in third countries where the output produced by the AI system is used in the Union. An AI agent calling, emailing or messaging people in the EU produces output used in the Union, which brings the transparency obligations into play.
What was deferred to 2027 — and what wasn’t?
The Digital Omnibus, Regulation (EU) 2026/1744, deferred the Annex III high-risk obligations from 2 August 2026 to 2 December 2027, and obligations for high-risk AI in Annex I products to 2 August 2028. Article 50 transparency was not deferred: it applies from 2 August 2026, with one transition — under the new Article 111(4), systems already on the market before that date have until 2 December 2026 to meet the Article 50(2) machine-readable marking duty. The Omnibus entered into force on 27 July 2026.
What happens if we ignore Article 50?
Non-compliance with the transparency obligations is expressly listed among the breaches subject to administrative fines under Article 99 of the AI Act, and the fines are significant. Member State authorities set enforcement within the Act’s framework, and factors like the nature, gravity and duration of the infringement affect the amount. Beyond fines, undisclosed AI is a reputational risk with prospects and a contractual risk with partners.
Updated 12 October 2026: re-verified Article 50 and Article 113 against the Official Journal text and the Commission’s pages; added all seven paragraphs of Article 50 with their exceptions, the full text of Article 50(4), the Article 3(60) deep fake definition, the Omnibus’s Article 111(4) transition and replacement Article 50(7), and the new Article 5 prohibitions applying from 2 December 2026. No change to any date. This article is general information for practitioners, current as at 12 October 2026, and is not legal advice. Regulations and guidance change; verify the current state at EUR-Lex and the European Commission, and speak to qualified counsel about your specific deployment.