Artificial intelligence has quietly become part of everyday interaction — from customer service chatbots to AI-generated marketing images and voice assistants. As these tools blend more seamlessly into daily life, one question has become urgent for regulators and businesses alike: how does a person know when they're dealing with a machine rather than a human, or when the image, video, or text in front of them was never created by a person at all?
That is precisely the gap that Article 50 of the EU AI Act was written to close. As of 2 August 2026, this provision is no longer a future compliance deadline — it is live, enforceable law across all 27 EU member states. Whether you run a customer-facing chatbot, build generative AI tools, or publish AI-assisted content, understanding Article 50 of the AI Regulation is now essential to operating legally in the European market.
This post breaks down what Article 50 actually requires, who it applies to, when it took effect, and how organizations are expected to comply — based on the official text of the Regulation and the European Commission's guidance.
This article covers:
What Article 50 of the EU AI Act is and why transparency in AI is now legally required
When Article 50 came into effect (2 August 2026) and what deadlines businesses must meet
The four core transparency obligations under Article 50 (AI interaction, content labeling, biometrics, deepfakes)
Disclosure rules for chatbots, virtual assistants, and AI-human interactions
The requirement to mark AI-generated content (text, image, video, audio) in a machine-readable format
Special rules for emotion recognition and biometric categorization systems
Deepfake disclosure requirements and exceptions for editorially reviewed content
Who must comply: providers vs. deployers, including scope for non-EU companies
Penalties for non-compliance (up to €15M or 3% of global turnover) and enforcement authorities
Practical compliance steps: AI audits, disclosures, labeling systems, and Code of Practice adoption
What Is Article 50 of the EU AI Act?
Article 50 sits within Chapter IV of Regulation (EU) 2024/1689, titled "Transparency Obligations for Providers and Deployers of Certain AI Systems." Unlike many other parts of the AI Act, it does not depend on whether a system is classified as "high-risk." Instead, it applies horizontally to any AI system that falls into one of four specific use cases — regardless of the sector, size of the company, or purpose of the tool.
According to the official AI Act Service Desk text of Article 50, the provision exists to reduce what regulators call "transparency risks"—impersonation, deception, manipulation, and misinformation—by making sure people know when they are interacting with AI, or being exposed to content that AI created or altered.
The European Commission has reinforced this rationale in its own guidance, noting that the goal is to ensure individuals are aware when they are interacting with AI systems or exposed to artificially generated or manipulated content, according to the Commission's guidelines published on Shaping Europe's Digital Future.
The Four Transparency Obligations Under Article 50
Article 50 is built around four distinct duties, each targeting a different transparency risk. It's important to understand that these obligations fall on different actors—sometimes providers, sometimes deployers, and in some cases, both.
1. Disclosure of AI Interaction (Article 50(1))
Providers must design AI systems intended to interact directly with people — chatbots, virtual assistants, voice bots, and similar tools — so that users are informed they are communicating with a machine, not a human. This obligation doesn't apply if it would already be obvious to a "reasonably well-informed" person that they are interacting with AI.
The Commission's FAQ clarifies that for this obligation to apply, the interaction must involve a genuine two-way exchange rather than passive data collection, must occur directly between the AI system and the person (not through a human intermediary), and must involve a natural person as the counterpart. Systems that operate purely in the background or communicate machine-to-machine fall outside this rule, per the Commission's official FAQ on transparency obligations.
2. Marking of AI-Generated Content (Article 50(2))
This is the labeling requirement under Article 50 of the AI Regulation that has generated the most industry attention. Providers of AI systems—including general-purpose AI systems—that generate synthetic audio, image, video, or text must ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated.
Crucially, the obligation requires that technical marking solutions be effective, interoperable, robust, and reliable, taking into account the type of content, implementation costs, and the state of the art reflected in relevant technical standards. The duty does not apply where AI performs only an assistive editing function or does not substantially alter input content, nor where content is used for law-enforcement purposes, as set out directly in the Article 50 text on artificialintelligenceact.eu.
3. Disclosure for Emotion Recognition and Biometric Categorization (Article 50(3))
Deployers using emotion recognition systems or biometric categorization systems must inform the people exposed to these systems about how they operate and must process any related personal data in line with the GDPR and other applicable EU data protection laws. An exemption applies to systems that are legally authorized for detecting, preventing, or investigating criminal offenses, subject to appropriate safeguards.
This duty is distinct from the outright ban on emotion recognition in workplaces and educational settings under Article 5 of the AI Act, which already applies. Outside those specific contexts, emotion recognition is generally permitted — but only with disclosure.
4. Disclosure of Deepfakes and AI-Generated Public Interest Text (Article 50(4))
Deployers who generate or manipulate image, audio, or video content that constitutes a "deepfake" — content that falsely appears authentic or truthful—must clearly disclose that the content is artificially generated or manipulated. Similarly, deployers who publish AI-generated or manipulated text on matters of public interest must disclose this, unless the content has undergone human review and an individual or organization holds editorial responsibility for it. Standard newsroom editorial review is generally sufficient to satisfy this exception; publishing unedited AI output directly is not.
When Does Article 50 of the AI Regulation Take Effect?
Article 50 of the AI Regulation formally entered into application on 2 August 2026. This date was set by the AI Act itself and confirmed by the European Commission, which notes its transparency guidelines were adopted specifically to support obligations that started to apply on that date, as detailed in the Commission's official guidelines page.
There is one important nuance for the machine-readable marking requirement under Article 50(2). Under the AI Omnibus provisional agreement reached in May 2026, generative AI systems that were already on the market before 2 August 2026 were granted an extended runway—until 2 December 2026—to fully implement the machine-readable marking requirement, as reported by artificialintelligenceact.eu's practical guide to Article 50. Outside of that narrow technical grace period, there is no general delay: disclosures under Article 50(1), (3), and (4) must be provided no later than the first interaction or exposure, with no allowance for informing someone after the fact.
It's also worth noting that Article 50 obligations are separate from — and unaffected by — the later deferral of the AI Act's high-risk system rules (Annex III) to December 2027 under the Digital Omnibus Regulation (EU) 2026/1744. That deferral applies to a different part of the Act entirely; the transparency rules under Article 50 remain on their original 2 August 2026 timeline.
Who Must Comply: Scope of Article 50
Article 50 casts an unusually wide net. It applies to:
Providers — organizations that develop, build, or place AI systems on the market or put them into service
Deployers — organizations that use AI systems under their own authority, including companies that integrate third-party AI tools into their products or workflows
Because the obligations attach to specific use cases rather than a risk classification, they reach far beyond systems formally labeled "high-risk." Chatbots, generative AI writing tools, image generators, voice synthesis tools, and AI-powered customer service platforms can all fall within scope. Notably, the obligations also extend to providers and deployers of open-source AI systems, which are not exempt from Article 50 despite receiving lighter treatment elsewhere in the Act.
The territorial reach is broad as well. Non-EU providers and deployers can fall under Article 50 where their AI outputs are used within the EU, although incidental, unforeseeable, or unauthorized downstream use is not, on its own, enough to trigger obligations for third-country providers, according to the National Law Review's summary of the Commission's final guidelines.
The Code of Practice on the Transparency of AI-Generated Content
To help businesses translate Article 50's legal text into practical action, the EU AI Office coordinated the development of the Code of Practice on the Transparency of AI-Generated Content. This voluntary instrument, finalized in June 2026 by independent experts through a multi-stakeholder process, operationalizes two specific duties: the marking obligation on providers under Article 50(2) and the labeling obligation on deployers under Article 50(4), as explained by Bird & Bird's analysis of the final Code.
Signing the Code is not mandatory, and adherence does not by itself constitute conclusive proof of legal compliance. However, it offers real practical value: signatories benefit from a more predictable, Commission-endorsed compliance pathway, and market surveillance authorities are expected to treat the Code as a benchmark for good practice during enforcement. Organizations that choose not to sign must instead demonstrate compliance through other means the Commission considers adequate—a path with far less certainty attached to it.
By the end of July 2026, roughly 190 organizations across sectors, including IT, telecom, education, and retail had signed the Code ahead of the 2 August deadline, according to the Commission's own announcement on Shaping Europe's Digital Future. Signatories are also invited to join task forces launching in September 2026 to share implementation best practices and refine marking techniques over time.
Pending a unified EU-wide labeling icon, the Code allows signatories to rely on interim two-letter markers—such as "AI," "KI," or "IA"—while committing to adopt a common interactive EU icon once one is finalized.
Exemptions Under Article 50
Article 50 includes several carve-outs to avoid imposing disproportionate burdens:
Law enforcement exception: Systems authorized by law to detect, prevent, investigate, or prosecute criminal offenses are generally exempt from disclosure duties, subject to safeguards for third-party rights.
Assistive editing exception: The marking obligation under Article 50(2) does not apply where AI performs a standard assistive editing function or does not substantially alter the semantics of deployer-provided input.
Obvious-to-a-reasonable-person exception: The chatbot disclosure duty does not apply where AI interaction would already be evident to a reasonably well-informed person.
Editorial review exception: Disclosure of AI-generated public-interest text is not required where the content has been reviewed by a human and an entity holds editorial responsibility for publication.
Enforcement and Penalties
Non-compliance with Article 50 is not a minor administrative matter. Under Article 99 of the AI Act, breaches of the transparency obligations under Article 50 can result in administrative fines of up to €15 million or 3% of a company's total worldwide annual turnover for the preceding financial year, whichever is higher — as set out directly in Article 99 on the AI Act Service Desk.
Enforcement responsibility is split across three bodies. National market surveillance authorities in each member state handle the bulk of cases; the EU AI Office takes on systems that fall under its own direct supervision (typically involving general-purpose AI models); and the European Data Protection Supervisor steps in when an EU institution itself acts as a provider or deployer.
Practical Steps for Compliance
Organizations preparing for or catching up with Article 50 obligations should consider the following actions:
Map every AI touchpoint. Identify all chatbots, virtual assistants, content generators, and biometric or emotion-recognition tools used across your organization, including third-party and open-source tools.
Determine your role. Establish whether your organization acts as a provider, a deployer, or both for each system — the obligations differ accordingly.
Build disclosure into the user experience. Labels and disclosures must be clear, distinguishable, and accessible, delivered no later than the first interaction — buried disclaimers in terms and conditions or footers are unlikely to satisfy the requirement.
Implement machine-readable marking. For generative AI outputs, adopt watermarking, metadata standards (such as C2PA Content Credentials), or other detectable marking techniques appropriate to the content type.
Evaluate the Code of Practice. Signing the Code of Practice on the Transparency of AI-Generated Content can simplify demonstrating compliance and reduce regulatory uncertainty.
Establish editorial workflows. For AI-assisted public-interest content, ensure genuine human review and clear editorial accountability to qualify for the relevant exemption.
Document everything. Maintain records of compliance measures, as authorities are expected to weigh good-faith efforts and documented processes when assessing penalties.
Final Thoughts
The article in the AI Act Article 50 provides arguably the most wide-reaching transparency requirements yet imposed upon artificial intelligence, not least due to the fact it mandates you inform the humans the systems interacts with prior to them even having to decide they are high-risk systems. Any business either building or using chatbots, generative AI-generated content, or biometric systems affecting any EU based user cannot take a back seat any longer; that's their actively regulated role from August 2 2026 onward. Get ahead of the curve, have your explanations, your content marking and adopt practices from the Code of Practice – much better positioning than when the authorities start taking action.
FAQs
What is Article 50 of the EU AI Act?
It is the provision in the EU AI Act that sets transparency requirements for AI systems, requiring disclosure when people interact with AI, marking of AI-generated content, and disclosure of deepfakes and emotion-recognition or biometric categorisation systems.
When does Article 50 of the AI Regulation take effect?
Article 50 took effect on 2 August 2026, with a limited extension to 2 December 2026 for the machine-readable marking requirement under Article 50(2) for generative AI systems already on the market before the general deadline.
Does Article 50 apply only to high-risk AI systems?
No. Article 50 applies to any AI system used in the four situations it covers, regardless of whether that system is separately classified as high-risk.
What is the labeling requirement under Article 50 of the AI Regulation?
Providers of generative AI systems must mark synthetic audio, image, video, or text outputs in a machine-readable, detectable format, using solutions that are effective, interoperable, robust, and reliable.
Is signing the Code of Practice on the Transparency of AI-Generated Content mandatory?
No, it is voluntary. However, it offers a recognized, streamlined pathway to demonstrate compliance with the marking and labeling duties under Article 50(2) and (4).
What are the penalties for non-compliance with Article 50?
Fines can reach up to €15 million or 3% of global annual turnover, whichever is higher, under Article 99 of the AI Act.