European Union / Regulation (EU) 2024/1689
EU AI Act Article 50: transparency obligations
Article 50 makes providers tell people they are interacting with an AI and mark AI-generated content, and makes deployers disclose deepfakes, some AI-written text, and emotion recognition or biometric categorisation. It has applied since 2 August 2026. Generative systems already on the market before that date have until 2 December 2026 for the marking duty only.
01 / The requirements
Who owes what: providers and deployers
Article 50 binds two roles defined in Article 3. A provider develops an AI system, or has one developed, and places it on the market or puts it into service under its own name or trademark, paid or free (Article 3(3)). A deployer uses an AI system under its authority, except in a personal non-professional activity (Article 3(4)). Start with your role, then the content type, then the exceptions.
An organisation can be both. A company that builds its own generative system and uses it to make deepfakes must mark the output under 50(2) and label the deepfakes under 50(4) (Commission guidelines, para 15). Location does not exempt you: the Act covers providers and deployers outside the EU when the system's output is used in the Union (Article 2(1)(c)).
02 / Find your duty
Quick decision
Each branch comes from the duties below. Read the paragraph before relying on it; every duty has its own exceptions.
IF YOUprovide a chatbot, AI agent or other system people interact with directly
Tell them it's AI, unless that's obvious
§50(1)IF YOUprovide a system that generates synthetic audio, images, video or text
Mark outputs in a machine-readable, detectable format
§50(2)IF YOUplaced that generating system on the market before 2 Aug 2026
Mark by 2 Dec 2026; the other duties apply from 2 Aug 2026
§50(2)IF YOUuse AI to generate or manipulate a deepfake
Disclose it; for evidently artistic, creative, satirical or fictional work, without hampering the work
§50(4)IF YOUpublish AI-generated text to inform the public on matters of public interest
Disclose it, unless a human reviewed it and someone holds editorial responsibility
§50(4)A hosting service, online platform or broadcaster that only disseminates AI content made by others, without authority over the AI system used to make it, is not a deployer (guidelines, para 16). Neither is a person using AI for a purely personal, non-professional activity (Article 3(4)). The trap: a company that modifies someone else's generative AI system, for example with new training data, and puts it into service under its own name becomes the provider of the new system (guidelines, para 11).
03 / Dates & transition
Two dates, kept apart
Every Article 50 duty applies from this date, to every system in scope, whenever it was placed on the market, with one exception: the 50(2) marking duty for generative systems already on the market, below (Article 113; guidelines, para 153).
Providers of generative AI systems, including general-purpose AI systems, placed on the market before 2 August 2026 must meet the 50(2) marking duty by this date (Article 111(4)).
Article 50 did not move. The Digital Omnibus on AI, Regulation (EU) 2026/1744, in force since 27 July 2026, moved the high-risk obligations to 2 December 2027 (Annex III) and 2 August 2028 (Annex I). Inside Article 50 it rewrote only paragraph 7, on codes of practice. Around it, it added the four-month grace period for 50(2) marking in a new Article 111(4), made the AI Office exclusively competent for some systems (Article 75(1)), and capped fines for small mid-caps (Article 99(6a)).
04 / The duties
Each duty, with its exceptions
The requirement is only the starting point. Read each duty with its exceptions, rather than treating a label as the whole answer. The quoted text is the consolidated Regulation as it stands after the Digital Omnibus.
Provider
Article 50(1): telling people they are interacting with an AI system, such as a chatbot
Providers of AI systems intended to interact directly with natural persons. The Commission gives chatbots, AI agents and avatars as examples (Commission Q&A).
Design and develop the system so the people concerned are informed that they are interacting with an AI system.
Where it is obvious to a reasonably well-informed, observant and circumspect person, given the circumstances and context of use. The guidelines read this restrictively: general awareness that chatbots exist does not mean people recognise one in an interaction (para 45). Also excepted: systems authorised by law to detect, prevent, investigate or prosecute criminal offences, unless the public can use them to report a crime.
▸ Read Article 50(1) as written
Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting 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. This obligation shall not apply to AI 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 those systems are available for the public to report a criminal offence.
Provider
Article 50(2): machine-readable marking of AI-generated output, including text
Providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text.
Mark outputs in a machine-readable format so they are detectable as artificially generated or manipulated. The technical solution must be effective, interoperable, robust and reliable as far as technically feasible.
Systems that perform an assistive function for standard editing, systems that do not substantially alter the input data or its meaning, and uses authorised by law for criminal investigations. The Commission also places some outputs outside the duty, such as short sequences of numbers, symbols or letters, source code, and outputs exchanged only machine to machine (Commission Q&A).
▸ Read Article 50(2) as written
Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated. Providers shall ensure their technical solutions are effective, interoperable, robust and reliable as far as this is technically feasible, taking into account the specificities and limitations of various types of content, the costs of implementation and the generally acknowledged state of the art, as may be reflected in relevant technical standards. This obligation shall not apply to the extent the AI systems perform an assistive function for standard editing or do not substantially alter the input data provided by the deployer or the semantics thereof, or where authorised by law to detect, prevent, investigate or prosecute criminal offences.
Deployer
Article 50(3): telling people an emotion recognition or biometric categorisation system is running
Deployers of an emotion recognition system, which identifies or infers emotions or intentions from biometric data (Article 3(39)), or a biometric categorisation system, which assigns people to categories from biometric data (Article 3(40)).
Inform the people exposed that the system is operating, and process their personal data under the GDPR, Regulation (EU) 2018/1725 or Directive (EU) 2016/680 as applicable. The Commission says this applies whether the system runs in real time or after the fact, and does not require explaining its purpose (Commission Q&A).
Systems permitted by law to detect, prevent or investigate criminal offences, with safeguards and in line with Union law.
▸ Read Article 50(3) as written
Deployers of an emotion recognition system or a biometric categorisation system shall inform the natural persons exposed thereto of the operation of the system, and shall process the personal data in accordance with Regulations (EU) 2016/679 and (EU) 2018/1725 and Directive (EU) 2016/680, as applicable. This obligation shall not apply to AI systems used for biometric categorisation and emotion recognition, which are permitted by law to detect, prevent or investigate criminal offences, subject to appropriate safeguards for the rights and freedoms of third parties, and in accordance with Union law.
Deployer
Article 50(4): disclosing deepfakes
Deployers of an AI system that generates or manipulates image, audio or video content constituting a deepfake: content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful (Article 3(60)).
Disclose that the content has been artificially generated or manipulated. The Commission says the label must be perceivable without technical tools, so a deployer cannot rely on the provider's machine-readable mark under 50(2) alone (Commission Q&A).
Uses authorised by law for criminal investigations. For evidently artistic, creative, satirical, fictional or analogous work, the duty shrinks to disclosing that such content exists, in a way that does not hamper the display or enjoyment of the work.
▸ Read Article 50(4) as written
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.
Deployer
Article 50(4): disclosing AI-generated text published to inform the public on matters of public interest
Deployers of an AI system that generates or manipulates text published with the purpose of informing the public on matters of public interest.
Disclose that the text has been artificially generated or manipulated.
Uses authorised by law for criminal investigations, or text that has undergone human review or editorial control where a natural or legal person holds editorial responsibility for the publication. Both conditions must hold. Spell-checks and other superficial checks do not count as review (Commission Q&A), and any substantive AI change after editorial sign-off voids the exception (guidelines, para 136). The exception sits in paragraph 4, so it does not remove the provider's separate marking duty under 50(2).
▸ Read Article 50(4) as written
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.
Provider and deployer
Article 50(5): how and when the information is given
Clearly and distinguishably, at the latest at the first interaction or exposure, and in line with applicable accessibility rules. The guidelines say Article 50 adds no accessibility requirements of its own (para 144).
Every notice and label under paragraphs 1 to 4.
▸ Read Article 50(5) as written
The information referred to in paragraphs 1 to 4 shall be provided to the natural persons concerned in a clear and distinguishable manner at the latest at the time of the first interaction or exposure. The information shall conform to the applicable accessibility requirements.
▸ 50(6) and 50(7): what the article does not change, and the codes of practice
ARTICLE 50(6)
Paragraphs 1 to 4 do not affect the high-risk requirements in Chapter III, and other Union or national transparency rules for deployers still apply. Recital 137 adds that meeting Article 50 does not mean the system or its output is lawful.
ARTICLE 50(7), AS REWRITTEN BY THE DIGITAL OMNIBUS
The Commission encourages codes of practice on detection, marking and labelling, and assesses, taking utmost account of the AI Board's opinion, whether adherence is adequate for paragraphs 2 and 4. If it finds a code inadequate, it may adopt an implementing act with common rules.
05 / Guidelines & Code of Practice
Commission guidelines and the Code of Practice
Two documents sit beside the text. Neither replaces it. The Commission's guidelines explain scope and exceptions for all of Article 50. The Code of Practice is a voluntary way to show compliance with the marking and labelling duties in 50(2), (4) and (5).
Published 20 July 2026. The English text now on the Commission's page is Communication C(2026) 7702 final, dated 5 November 2026. It covers who is a provider or deployer, the four tests for an interactive system, what counts as a deepfake, standard editing, public-interest text, and enforcement.
Drawn up by independent experts in a process the AI Office facilitated. Section 1 is for providers (marking and detection, 50(2) and (5)); Section 2 is for deployers (labelling deepfakes and text, 50(4) and (5)). The EU also published optional icons for labels. About 190 organisations had signed by the end of July 2026.
On 8 July 2026 the Commission concluded the Code adequately covers Articles 50(2), (4) and (5); the AI Board adopted its own adequacy assessment the next day. Adherence still does not constitute conclusive evidence of compliance.
▸ What the Code asks of signatories, and what happens if you don't sign
MARKING
As long as no single technique can meet all four requirements of 50(2) on its own, signatories mark audio, images, video and containerised text with at least two machine-readable layers: digitally signed metadata and an imperceptible watermark (Code, Measure 1.1). Free-form text, which cannot carry metadata, needs only the watermark layer.
IF YOU SIGN
Authorities focus on whether you adhered to the Code and implemented its measures (guidelines, para 147). Commitments implemented under an adequate code can count as a mitigating factor when fines are set (para 149). Opting out of a section loses that benefit for the section.
IF YOU DON'T
You must show compliance with 50(2), (4) and (5) by other adequate means, for example with a gap analysis against the Code, and you may face more requests for information (para 148).
50(1) AND 50(3)
No code covers these. Providers and deployers choose their own measures, taking the guidelines into account (Commission Q&A).
HOW IT GOT HERE
First draft of the Code 17 December 2025; second draft 3 March 2026; draft guidelines put to consultation 8 May 2026; final Code 10 June 2026. The AI Office will consider facilitating formal updates to the Code at least every two years.
06 / Timeline, enforcement & fines
How the dates fit together
The AI Act enters into force. Regulation (EU) 2024/1689, with duties phased in over the following years.
Final Code of Practice published. Found adequate by the Commission on 8 July 2026.
Commission guidelines on Article 50 published. The current English text is C(2026) 7702 final of 5 November 2026.
The Digital Omnibus enters into force. High-risk dates move to 2027 and 2028. Article 50 keeps its date.
Article 50 applies to every system in scope. Only the 50(2) marking duty for generative systems already on the market waits until 2 December.
The marking grace period ends. Generative systems placed on the market before 2 August 2026 must meet 50(2).
National market surveillance authorities enforce most of Article 50. The AI Office is exclusively competent for AI systems built on a general-purpose AI model when one provider, or one undertaking, makes both, and for AI systems that are or sit inside a designated very large online platform or search engine, with some exceptions (Article 75(1)). That competence covers the providers of those systems, and their deployers only when the deployer is the provider or part of its undertaking; other deployers answer to national authorities. The European Data Protection Supervisor covers EU institutions.
Of total worldwide annual turnover for the preceding financial year, whichever is higher. For SMEs and start-ups, and for small mid-caps, whichever is lower (Article 99(6) and (6a)). EU institutions face up to €750,000 (Article 100(3)). This is not the Act's ceiling: prohibited practices reach €35,000,000 or 7%.
Anyone with grounds to think Article 50 has been breached can complain to a market surveillance authority (Article 85; guidelines, para 151). Fines weigh the factors in Article 99(7), including the nature, gravity and duration of the breach and the operator's cooperation.
07 / Optional evidence · where Planisphere fits
Your team takes action. The evidence travels with it.
Your team presents the notice or applies the label. Planisphere seals the event your team reports, as a record anyone can check later.
Your interface tells the person they are interacting with AI, or labels the content.
The event your team reports is sealed when it happens, so a later edit shows.
Paste the receipt into Verify. No account or key.
A record is not a legal conclusion. It doesn't establish that a disclosure was made or that Article 50 is satisfied.
What a receipt establishes08 / Questions
The questions that come next
What does Article 50 of the EU AI Act require?+
Transparency in four cases. Providers must tell people when they are interacting with an AI system (50(1)) and mark AI-generated audio, images, video and text in a machine-readable format (50(2)). Deployers must inform people exposed to emotion recognition or biometric categorisation (50(3)), and disclose deepfakes and AI-generated text published to inform the public on matters of public interest (50(4)). Paragraph 5 sets the manner: clear, distinguishable, and no later than the first interaction or exposure.
When does Article 50 apply?+
From 2 August 2026, for every system in scope. Providers of generative AI systems placed on the market before that date have until 2 December 2026 to meet the 50(2) marking duty (Article 111(4)). That grace period does not cover the other transparency duties.
Does Article 50 apply to chatbots?+
Yes, under Article 50(1). The provider of a chatbot, AI agent or other system meant to interact directly with people must design it so they are told they are interacting with an AI system, unless that is obvious to a reasonably well-informed, observant and circumspect person in that context. The Commission reads the exception restrictively: general awareness that chatbots exist does not mean people recognise one in an interaction.
What counts as a deepfake?+
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. The Commission's Q&A reads this as three tests that must all be met (the guidelines split the same definition into four): resemblance, something that exists or plausibly could, and a false appearance of authenticity. Under this definition text cannot be a deepfake; AI-written text has its own rule in the second part of 50(4).
Is C2PA enough?+
Neither the Code of Practice nor the Commission guidelines name C2PA. Under the Code, signatories mark audio, images, video and containerised text with at least two machine-readable layers, digitally signed metadata and an imperceptible watermark, so a metadata standard alone covers one layer. Read the full answer.
Does Planisphere certify compliance?+
No. A record can support an evidence process; it is not a legal finding or a certification.
Our system launched last year. Do we have until December?+
For the marking duty only. Article 111(4) gives providers of generative AI systems placed on the market before 2 August 2026 until 2 December 2026 to comply with Article 50(2). The interaction disclosure, the deployer duties, and the manner-and-timing rule have applied since 2 August. A system that is partly interactive and partly generative gets the extension for its marking and not for its disclosure (guidelines, para 153).
Do we have to go back and mark everything we generated before?+
No. Outputs and deepfakes generated before 2 August 2026 do not need to be marked or labelled retroactively, though the Commission encourages deployers to label existing content where possible. One asymmetry: text generated before that date and published on or after it does need to be labelled (guidelines, para 154).
We signed the Code. Are we covered?+
You have a recognised way to show compliance with 50(2), (4) and (5): authorities will focus on whether you adhered to the Code, and commitments implemented under it can be a mitigating factor at the fining stage. You do not have a presumption of conformity. The Digital Omnibus says codes under 50(7) do not grant one (recital 41), and the Commission says adherence is not conclusive evidence of compliance. What signing changes is where authorities focus: from whether your marking is adequate to whether you adhered to the Code (guidelines, para 147).
Does Article 50 apply to us from outside the EU?+
Yes, if you place the system on the EU market, or if its output is used in the Union (Article 2(1)(a) and (c)). The guidelines add that incidental, unforeseeable or unauthorised downstream use should not alone trigger the duties for a non-EU provider whose system is not placed on the market or put into service in the Union (para 10).
Who enforces Article 50, and what are the fines?+
Mostly national market surveillance authorities; the AI Office for the providers of systems built on a general-purpose AI model by the same provider or undertaking, and of AI in designated very large online platforms and search engines (Article 75(1)). Breaches of Article 50 carry fines up to €15,000,000 or 3% of worldwide annual turnover, whichever is higher, and whichever is lower for SMEs and small mid-caps (Article 99).
How does this compare with California?+
Read the two laws separately. Article 50 splits its duties between providers and deployers, and its text names no specific data elements for the marking. California's rules are on their own page: California SB 942 disclosure and detection requirements.
09 / Primary sources
Go back to the source
Read the primary text alongside the guidance. Each source has a defined role in this explanation, and each was read for this page on 12 November 2026.
Article 50 as it reads now, with the Digital Omnibus changes marked. The Official Journal texts remain authoritative.
Article 50; the definitions in Article 3(3), (4), (39), (40) and (60); penalties in Articles 99 and 100; dates in Article 113; recitals 132 to 137.
Signed 8 July 2026, published 24 July 2026, in force 27 July 2026. Adds Article 111(4), rewrites Article 50(7), and moves the high-risk dates.
Published 20 July 2026; the English text is C(2026) 7702 final of 5 November 2026. Paragraph numbers on this page refer to it.
Provider and deployer roles, the deepfake tests, public-interest text, the grace period and enforcement.
Final text of 10 June 2026, its drafting timeline, and how to sign.
8 July 2026: the Code adequately covers Articles 50(2), (4) and (5); with the AI Board's assessment of 9 July.
Harmonised standards work for the AI Act.
10 / Change log
What changed, and when
Every statement rechecked against the consolidated text, the Digital Omnibus, the Commission guidelines (C(2026) 7702) and the Code of Practice. Added a section per duty with its exceptions, the guidelines and Code section, and fines for SMEs, small mid-caps and EU institutions. Provision quotes now match the official wording. Removed a 2 February 2027 Code deadline and a claim about fine-tuning that the official texts do not support.
Layout revision published. This revision does not imply a new review of every source.
Scoped accuracy corrections reviewed.
Summary and two introductory answers checked against Commission guidance.
Related: AI labeling laws compared · California SB 942 · China labelling Measures · India IT Rules · Korea AI Framework Act · EU AI Act: dates, bans and fines