BuildNot yet confirmed elsewhere1 publisher2 min readPublished Updated
EU AI Act Article 50 gives generators already on the market until 2 December to mark output
EU AI Act Article 50 has applied since 2 August 2026 to chatbots and synthetic media, high-risk or not, according to a dev.to checklist. Generative systems already on the EU market before that date have until 2 December 2026 to add machine-readable marking.
The Engineer · Build desk
Timeline: 27 July 2026, Digital Omnibus on AI in force. 2 August 2026, Article 50 applies. 2 December 2026, marking duty for generative systems already on the EU market. 2 December 2027, Annex III high-risk duties. 2 August 2028, Annex I regulated products.
- 2026Digital Omnibus on AI in force; high-risk duties postponed, claim 7
- 2026Article 50 transparency duties apply; not postponed, claim 8
- 2026Machine-readable marking due for generators already on EU market, claim 9
- 2027Annex III high-risk obligations apply, claim 7
- 2028Annex I: AI embedded in regulated products, claim 7
| Date | Event | Claim |
|---|---|---|
| July 27, 2026 | Digital Omnibus on AI in force; high-risk duties postponed | 7 |
| August 2, 2026 | Article 50 transparency duties apply; not postponed | 8 |
| December 2, 2026 | Machine-readable marking due for generators already on EU market | 9 |
| December 2, 2027 | Annex III high-risk obligations apply | 7 |
| August 2, 2028 | Annex I: AI embedded in regulated products | 7 |
What happened
- Article 50(1) puts a design duty on providers: systems that interact with people must tell users they are dealing with AI, unless that is obvious from context.
- Article 50(2) requires providers of systems that generate synthetic audio, images, video or text to mark outputs in a machine-readable format so they can be detected as AI-generated or manipulated.
- Under Article 50(4), deployers carry the disclosure duty for deepfakes and for AI-written text put out to inform readers on public-interest topics.
- The Digital Omnibus on AI, in force since 27 July 2026, pushed Annex III high-risk obligations to 2 December 2027 and Annex I products to 2 August 2028.
- Article 50(5) requires disclosures to be clear, distinguishable and accessible, and given at the latest at first interaction or exposure.
Why it matters
- decision Teams that both build and publish have to assign provider and deployer duties feature by feature, because one product can hold both roles.
- constraint Owners of generators already on the market have a fixed 122-day window to build and test marking that third parties can detect.
- cost Skipping the label on published AI text about public-interest topics depends on a documented human review and a named editorial owner, so someone has to keep the review record.
The two headline duties attach to different parts of a product. Article 50(1) is an interface duty. The checklist asks for a clear notice at or before the first message and allows a skip only where context makes AI use obvious, with the reason documented [14]. For audio-only content it calls for an audio disclaimer [15].
Article 50(2) attaches to the output. The checklist says to embed watermarks, metadata or equivalent signals and to aim for solutions that are effective, interoperable, robust and reliable as far as technically feasible [11]. The qualifier "as far as technically feasible" leaves the bar for "robust" open. The post does list what a mark has to survive: compression, cropping and format conversion, with detection that works for third parties [12]. In its words, "A mark nobody can read is a failed control." [13]
We'd expect a mark stored only in file metadata to fail the format-conversion test first. Assistive editing that does not substantially alter the input is carved out of the marking duty [19].
The Omnibus left Article 50 alone [8]. High-risk Annex III systems now trail it by 16 months [17], and a simple marketing chatbot gets no such lag [2].
The line between old and new products is where the exposure sits. Anything placed on the market from 2 August 2026 must comply from day one [10]. The checklist does not say whether a new model version of an existing product counts as a new placement. A team that swaps the model under a live product has to decide which side of the line it stands on.
The dates rest on that single post. No regulation text or Commission document sits alongside it to check them against.
What to watch
- Whether the regulation text or Commission guidance says when a new model version of an existing product counts as placed on the market.
- Any technical specification that fixes what robust marking means, since the checklist only says effective and reliable as far as technically feasible.
- 2 December 2026, when the marking duty reaches generative systems that were already on the EU market.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence45
- Adoption
- Insufficient
- Hype gap+15
- Incentives70
- Confidence50
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Article 50 of Regulation (EU) 2024/1689 sets transparency duties for AI systems that talk to people or create content.
ReportedSupportedSource: dev.to checklist, EU AI Act Article 50: AI Labelling Checklist for Developers (2026)View cited source - [2]
Article 50 applies whether or not a system is classed as high-risk, so even a simple marketing chatbot is in scope.
- [3]
Article 50(1): providers must design AI systems that interact with people so users are told they are dealing with AI, unless that is obvious from context.
- [4]
Article 50(2): providers of systems generating synthetic audio, images, video or text must mark outputs in a machine-readable format so they are detectable as AI-generated or manipulated.
- [5]
Article 50(4): deployers must disclose deepfakes, and AI-generated text published to inform the public on matters of public interest.
- [6]
Article 50(5): disclosures must be clear, distinguishable, accessible, and given at the latest at first interaction or exposure.
- [7]
The Digital Omnibus on AI, in force since 27 July 2026, postponed high-risk obligations: Annex III systems now follow on 2 December 2027, and AI embedded in regulated products (Annex I) on 2 August 2028.
- [8]
Article 50 was not postponed and has applied since 2 August 2026.
- [9]
For generative systems already on the EU market before 2 August 2026, the Article 50(2) machine-readable marking duty applies from 2 December 2026.
- [10]
Anything placed on the market from 2 August 2026 must comply from day one.
- [11]
The checklist says to embed watermarks, metadata or equivalent signals into generated audio, image, video and text, and to aim for solutions that are effective, interoperable, robust and reliable as far as technically feasible.
- [12]
The checklist says to test that marking signals survive compression, cropping and format conversion, and that detection works for third parties.
- [13]
A mark nobody can read is a failed control.
- [14]
The checklist says to show a clear 'you are talking to an AI' notice at or before the first message, skipping it only where context makes AI use genuinely obvious, and to document why.
- [15]
Audio-only content needs an audio disclaimer.
- [16]
From 2 August 2026 to 2 December 2026 is 122 days.
- [17]
Annex III high-risk obligations now apply 16 months after Article 50 does (2 December 2027 versus 2 August 2026).
- [18]
A SaaS company that fine-tunes a model and embeds it in its product is a provider for Articles 50(1) and 50(2), and a deployer for 50(4) when it publishes outputs itself; the checklist says to map the role per feature, not per company.
- [19]
Assistive editing that does not substantially alter the input is carved out of the marking duty.
- [20]
Published AI-generated text that informs the public needs a label, unless a human has reviewed it and a person holds editorial responsibility; the checklist says to record that review.
Sources
1 independent publisher whose own reporting we read for this story.
- dev.toEU AI Act Article 50: AI Labelling Checklist for Developers (2026)
1 article · October 12, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- AI regulation complianceFollow
- AI Content Labelling and DisclosureFollow
- EU AI ActFollow