Guide · EU AI Act Article 50
Marking AI-generated content by 2 December 2026.
If your product generates text, images, audio or video and was on the EU market before 2 August 2026, its outputs must be marked in a machine-readable way by 2 December 2026. This guide sets out who that applies to, what to check, and how to see where your own product stands.
Checked against the texts on EUR-Lex on 27 October 2026: Regulation (EU) 2024/1689 (AI Act) and Regulation (EU) 2026/1744 (Digital Omnibus on AI). General information, not legal advice.
What changes on 2 December 2026
2 August 2026
Article 50 applies. Chatbots and other systems that interact with people must tell them they are talking to an AI system, unless that is obvious. Generative systems placed on the market from this date must mark their outputs in a machine-readable way.
2 December 2026
The four-month transitional period ends. Providers of generative AI systems that were already on the market before 2 August 2026 must comply with the machine-readable marking duty in Article 50(2).
Regulation (EU) 2026/1744 added Article 111(4) to the AI Act: providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content that were placed on the market before 2 August 2026 must take the necessary steps to comply with Article 50(2) by 2 December 2026. New prohibited practices that the same regulation added to Article 5 also apply from 2 December 2026.
Who it applies to
The 2 December date concerns you when all three conditions hold.
01
You provide the system
The duty sits with the provider: the company that develops the system, or has it developed, and puts it on the market under its own name. A company that only uses another provider’s tool is a deployer and has different duties (see below).
02
It generates synthetic content
The system, including a general-purpose AI system, generates audio, image, video or text content.
03
It was on the EU market before 2 August 2026
Only these systems had the transitional period. A system placed on the market on or after 2 August 2026 had to mark its outputs from the start.
If you deploy someone else’s system
Deployers have their own Article 50 duties, which have applied since 2 August 2026: disclose deep fakes (image, audio or video), disclose AI-generated text published to inform the public on matters of public interest unless it has had human review or editorial control under someone’s editorial responsibility, and inform people exposed to emotion recognition or biometric categorisation. Whether you are a provider or a deployer depends on the facts; the same company can be both for different systems.
A practical checklist
The Act requires marking that is effective, interoperable, robust and reliable as far as technically feasible, taking into account the type of content, the costs of implementation and the generally acknowledged state of the art. It does not prescribe one technique; recital 133 names watermarks, metadata, cryptographic provenance methods, logging and fingerprints. Codes of practice under Article 50(7) and any implementing act the Commission adopts are the places to watch for more specific rules.
Article 50(1) and (5) · applies since 2 August 2026
Tell users they are talking to an AI system
- Show the notice in the interface at the latest at the first interaction, for example in the chat window before or with the first reply.
- Make it clear and distinguishable from the rest of the content, and meet the accessibility requirements that apply to you.
- Cover every entry point: web widget, mobile app, voice line, messaging integrations and email replies generated by the system.
- Keep a dated screenshot or the code that renders the notice, so you can show when and where it appears.
Article 50(2)
Mark generated text in a machine-readable way
- Add provenance metadata where the text leaves your system: API response fields, document properties of exported files, or headers.
- Consider a statistical text watermark from your model provider or an open library, and record its known limits on short or heavily edited text.
- Make sure the marking survives your own export and copy paths, not only the first response.
Article 50(2)
Mark generated images
- Attach content credentials (C2PA manifests) and set the IPTC digital source type in the file metadata.
- Add an invisible watermark where metadata is likely to be stripped, for example by social platforms or image resizing.
- Test that a public detection tool or your own verifier reads the mark after common transformations.
Article 50(2)
Mark generated audio and video
- Use an audio watermark for speech and music, and content credentials or metadata in the container file.
- For video, mark the file and consider frame-level watermarking for clips that are likely to be re-encoded.
- Document which outputs are marked, by which technique, and how you test detection.
Article 50(2), last sentence · recital 133
Decide whether “standard editing” covers a feature
- The marking duty does not apply where the system performs an assistive function for standard editing, or does not substantially alter the input data supplied by the deployer or its meaning.
- The Act does not list examples. Features that only correct spelling, grammar or formatting are the usual candidates; rewriting, summarising or generating new content usually are not.
- Write down, per feature, what the output changes compared with the input and why you consider it standard editing. Have that reasoning reviewed by your legal adviser.
Article 111(4), added by Regulation (EU) 2026/1744
Check the grace-period conditions
- Record the date the system was first placed on the EU market, with evidence such as a release note, a launch announcement or the first customer contract.
- If that date is before 2 August 2026, marking must be in place by 2 December 2026. If it is later, marking was already required.
- The transitional period concerns Article 50(2) only. The disclosure duties in Article 50(1), (3) and (4) applied from 2 August 2026.
Check your own product in Merron, free
Create a free account, no payment card required. Then:
- Read your code for Article 50 signals. Connect the read-only GitHub App and choose a repository. Merron reports, with file and line, user-facing AI disclosure text and content-marking libraries such as C2PA, SynthID or invisible watermarks. Finding them shows they appear in the code, not that they run in production.
- Confirm the facts. Answer whether you are the provider or deployer, what the system generates and when it was first placed on the EU market. Unknown answers stay unknown. The screening shows the Article 50 duties that follow and the date that applies to you.
- Draft the disclosure texts. Where an evidence review is enabled, Merron drafts Article 50 notices from the passages it found, each labelled as a draft for you to review and adopt.
Prefer your AI assistant? Connect Merron over MCP and run a check from selected code excerpts without an account.
Limits of this guide
This page summarises the legal texts as published on EUR-Lex on 27 October 2026. It is general information, not legal advice, and it does not cover national rules, sector rules or other Union law that may also require disclosure. Merron organises evidence and screens confirmed facts against versioned rules; it does not certify compliance or replace advice on your specific product.