Updated 12 November 2026 · First published 30 September 2026
2 December 2026 does not delay Article 50 chatbot disclosure, which has applied since 2 August 2026. It is a provider-only Article 50(2) marking transition for generative systems already on the market before 2 August (Article 111(4)). Two new Article 5 bans, on AI that generates or manipulates non-consensual sexual deepfakes or child sexual abuse material, also start that day.
Last checked: 10 November 2026, against the AI Act text on the Commission's AI Act Service Desk, the Commission's Article 50 guidelines and Q&A, the final Code of Practice on Transparency of AI-generated Content, the Commission's opinion on that code, and the C2PA 2.3 specification.
Key point: 2 December 2026 is a transition for one group and one duty: providers of generative AI systems already on the market before 2 August 2026, for Article 50(2) marking. If you run a chatbot or publish deep fakes, your Article 50 duties already apply.
Does 2 December 2026 delay Article 50 chatbot disclosure?
No: 2 December does not delay chatbot disclosure under Article 50(1), which has applied since 2 August 2026. Headlines about the Digital Omnibus "delaying the AI Act" are easy to over-read. The Omnibus delayed the high-risk rules, to 2 December 2027 and 2 August 2028. It did not move chatbot disclosure, deep-fake disclosure, or any other Article 50 duty, except one narrow, provider-only transition: Article 111(4) on the AI Act Service Desk gives providers of generative systems already on the market before 2 August 2026 until 2 December 2026 to meet Article 50(2) machine-readable marking. If you run a chatbot or publish deep fakes, those duties have applied since 2 August 2026. Paragraph 153 of the Commission's Article 50 guidelines, adopted 5 November 2026 (C(2026) 7702 final; non-binding), says a system that is partly interactive and partly generative gets the transition only for marking; its chatbot disclosure applied from 2 August 2026.
The two new Article 5 bans that also start on 2 December 2026 — on placing on the market, putting into service or use of AI that generates or manipulates non-consensual sexual deepfakes or child sexual abuse material, scoped by Article 5(1a) — are a separate matter, not an Article 50 delay. See was Article 50 delayed?.
What changes on 2 December 2026
On 2 December 2026, providers of generative AI already on the market before 2 August 2026 must meet Article 50(2) marking, and two new Article 5 bans on sexual deepfakes and CSAM start.
Article 50(2) transition (Article 111(4)). The Service Desk text of Article 111 reads: "Providers of AI systems, including general-purpose 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 Commission's Article 50 Q&A calls this "a limited grace period", only for older systems and only for the marking and detection duty.
New prohibitions (Article 5(1)(ba) and (bb)). Placing on the market, putting into service or using AI systems that generate or manipulate realistic images, video or audio of an identifiable person's intimate parts or of them engaged in sexually explicit activity, without their explicit consent, or child sexual abuse material, is banned from 2 December 2026 (Service Desk: Article 5). Article 5(1a) limits when placing such a system on the market or putting it into service is banned: broadly, where that output is its intended purpose, or is a reasonably foreseeable outcome and the system lacks adequate safeguards to prevent it. See the official text. These bans are an Article 5 matter, not Article 50, and the Article50.io scan doesn't check for them.
| Date | What applies | Source |
|---|---|---|
| 2 August 2026 | Article 50 transparency rules | Service Desk timeline |
| 2 December 2026 | Article 50(2) transition ends + new Article 5 bans | Art. 111(4); Service Desk timeline |
| 2 December 2027 | Annex III high-risk rules | Service Desk timeline |
| 2 August 2028 | Annex I high-risk rules | Service Desk timeline |
Article 50 dates, duty by duty
Every Article 50 duty has applied since 2 August 2026, except Article 50(2) marking for generative AI systems already on the market, which has until 2 December 2026.
- Article 50(1): tell people they are interacting with an AI system. Applies to providers from 2 August 2026. This is the chatbot, voice agent and AI assistant rule; for phone lines, see do AI phone and voice agents need a disclosure?
- Article 50(2): mark synthetic audio, images, video and text in a machine-readable format so it is detectable as AI-generated. Applies to providers of generative AI systems from 2 August 2026. Systems placed on the market before that date have until 2 December 2026.
- Article 50(3): inform people exposed to emotion recognition or biometric categorisation. Applies to deployers from 2 August 2026.
- Article 50(4): disclose deep fakes, and AI-generated text published to inform the public on matters of public interest. Applies to deployers from 2 August 2026.
For what each paragraph requires in practice, see Article 50 of the EU AI Act, explained.
What Article 50(2) asks providers to do
Article 50(2) requires providers of generative AI to mark outputs in a machine-readable, detectable way; under the Code of Practice that means signed metadata, such as a C2PA manifest, plus an imperceptible watermark.
The Act says providers must ensure outputs "are 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 this is technically feasible" (Service Desk: Article 50). It does not apply where the system performs an assistive function for standard editing or does not substantially alter the input.
No blanket retroactive marking. Paragraph 154 of the Commission guidelines (C(2026) 7702 final, 5 November 2026, non-binding) says outputs within Article 50(2), and deep fakes within Article 50(4), "which have been generated or manipulated before 2 August 2026 do not need to be marked or labelled retroactively". The same applies to public-interest text that was both generated and published before that date. Text generated before 2 August 2026 but first published on or after it still needs a label where Article 50(4) applies. The 20 July 2026 text (C(2026) 5054 final) was a draft submitted for approval; paragraph 154 said the same.
What the Code of Practice expects. The Code of Practice on Transparency of AI-generated Content, published in final form on 10 June 2026, is voluntary. Under its Measure 1.1, signatories currently use "at least two layers of machine-readable marking" for audio, images, video and containerised text: digitally signed metadata (Sub-measure 1.1.1) and an imperceptible watermark (Sub-measure 1.1.2). Because free-form text can't carry metadata, a watermark alone is treated as enough for it.
How the Commission assessed the code. In its opinion of 8 July 2026, the Commission concluded the code adequately covers Articles 50(2), (4) and (5); the AI Board adopted its own assessment the next day. The Commission adds that adherence "does not constitute conclusive evidence of compliance".
Where C2PA fits. C2PA is an open standard for signed provenance records (Content Credentials); version 2.3 is dated December 2025. For Article 50(2), a C2PA manifest can carry the signed-metadata layer, but it is not a watermark, so on its own it covers one of the code's two layers. Neither the AI Act nor the code names a required standard.
Key point: Marking is the generator provider's duty. If you only use AI tools, ask your vendors in writing which marking layers they apply, and check that your own resizing or re-exporting doesn't strip them.
What a live-site scan can and can't observe for Article 50(2)
A scan of a public web page sees what that page serves. It can't see how a generator marks its outputs.
What the Article50.io scan does. After the browser closes, the scan reads the start of up to 8 of the page's main images and checks their metadata for an IPTC digital source type or a C2PA (Content Credentials) manifest saying the image was AI-generated. It reports an image only if no label, caption, alt text or other text on the page says so (English wording only); a missing marker is never a finding. It also reads text inside chat-like frames among the first 25 frames on the page, for Article 50(1). The browser part of each scan is capped at 45 seconds, and nothing is clicked or opened.
What it can't observe. The Article 50(2) marking duty itself isn't observable from a public page. A missing marker on a published image doesn't show the provider failed to mark it, because metadata is often stripped by resizing or upload. A present marker doesn't show the provider's solution is effective, robust or reliable. The scan doesn't detect watermarks.
Your own generator. If you provide a generative AI system, any answer you give about its Article 50(2) marking is marked as declared by you and not checked by the scan. A clean scan doesn't mean you're compliant. See how the scan works.
Where the 2 December date comes from
The AI Act (Regulation (EU) 2024/1689) originally applied Article 50 in full from 2 August 2026. The Digital Omnibus on AI amended the Act and created two 2 December 2026 dates: a new Article 111(4), giving existing generative AI providers until then for Article 50(2), and new prohibitions in Article 5(1)(ba) and (bb).
Three things follow from the Article 111(4) wording:
- It is a provider rule. It gives companies that build generative tools extra time; businesses that only use those tools get none.
- It is about marking, not disclosure. Visible disclosure duties sit in Article 50(1) and 50(4), and those were not deferred. Paragraph 153 of the guidelines adds that a system that is partly interactive and partly generative gets the transition only for marking; its chatbot disclosure applies from 2 August 2026.
- It applies only to systems already on the market before 2 August 2026. A generator launched later owed marking from day one.
For the full Omnibus timeline, see was Article 50 delayed?
What to do before 2 December
For most websites, 2 December 2026 changes nothing: chatbot and deep fake disclosure already apply.
If you only use AI tools (a chat widget, an image generator, an AI writing tool):
- Don't treat 2 December as your deadline. Check your chatbot disclosure and any deep fake or public-interest text disclosures now. For common chat widgets, see how to add an AI disclosure to Intercom, Zendesk, Drift and Crisp.
- Ask your generative tool vendors whether outputs carry machine-readable marking, and whether your workflow strips it.
- Label deep fakes you publish. See deepfake disclosure.
If you build or sell a generative AI system:
- Work out whether you are the provider and when the system was first placed on the market.
- If it was on the market before 2 August 2026, plan machine-readable marking of its outputs by 2 December 2026. If it launched later, the duty already applies.
If you build or host an AI image, video or audio generator: check whether it could be used to create sexual deepfakes of real people or CSAM. The new Article 5 bans apply from 2 December 2026. This is outside Article 50; get legal advice.
Frequently asked questions
What is the 2 December 2026 AI Act deadline?
It is a provider-only Article 50(2) marking transition for generative AI systems already on the market before 2 August 2026 (Article 111(4)). It does not delay chatbot disclosure under Article 50(1), which has applied since 2 August. Two new Article 5 bans, on AI that generates or manipulates non-consensual sexual deepfakes or child sexual abuse material, also start that day.
Was Article 50 delayed to December 2026?
No. Article 50 has applied since 2 August 2026. The Digital Omnibus on AI added one narrow transition in Article 111(4): providers of generative AI systems already on the market before 2 August 2026 have until 2 December 2026 for Article 50(2) marking. Disclosure duties in 50(1), 50(3) and 50(4) did not move.
Does the 2 December 2026 deadline apply to my website chatbot?
No. Telling people they are talking to an AI is Article 50(1), which has applied since 2 August 2026. The Article 50 part of the 2 December date only gives extra time to providers of generative AI systems for machine-readable marking. A website using a chatbot should already show its disclosure.
Do images made before 2 August 2026 need marking?
No. Paragraph 154 of the Commission's guidelines (C(2026) 7702 final; non-binding) says Article 50(2) outputs and deep fakes made before 2 August 2026 need no retroactive mark or label. Same for public-interest text generated and published before that date. Later-published text still needs a label where Article 50(4) applies. Labelling older deep fakes is encouraged, without disproportionate efforts, not required.
Is C2PA enough for Article 50(2)?
Not on its own under the Code of Practice. The code currently expects two machine-readable layers for images, audio, video and containerised text: digitally signed metadata, which a C2PA manifest can carry, plus an imperceptible watermark. The AI Act itself names no specific technique, and signing the code is voluntary.
Can a website scan check Article 50(2) marking?
No. The marking duty belongs to the generator provider and isn't observable from a public page. The Article50.io scan reads IPTC and C2PA AI markers on sampled images only to flag a missing visible label. Any answer about your own generator's marking is marked as declared by you and not checked by the scan.
What are the fines for breaching Article 50?
Under Article 99(4)(g) of the AI Act, up to €15 million or 3% of total worldwide annual turnover, whichever is higher. For SMEs and start-ups the lower of the two applies (Article 99(6)), and Article 99(6a) applies the same lower cap to small mid-cap enterprises. Actual fines depend on the circumstances of each case.
Sources
- EC AI Act Service Desk – Timeline for the implementation of the EU AI Act
- EC AI Act Service Desk – Article 111 (AI Act Explorer)
- EC AI Act Service Desk – Article 50 (AI Act Explorer)
- EC AI Act Service Desk – Article 5 (AI Act Explorer)
- EC AI Act Service Desk – Article 99 (AI Act Explorer)
- EC AI Act Service Desk – FAQ: When does enforcement start?
- European Commission – Guidelines on transparency obligations: Commission Guidelines C/2026/6005 (OJ C, 12.11.2026), adopted as C(2026) 7702 final on 5 Nov 2026, non-binding (point 5); paras 153–154 · English PDF
- European Commission – Q&A: Transparency obligations under Article 50 of the AI Act
- European Commission – Code of Practice on Transparency of AI-generated Content
- European Commission – Opinion on the assessment of the Code of Practice (8 July 2026)
- C2PA Technical Specification 2.3
- European Parliament Legislative Train – Digital Omnibus on AI
This article is general information, not legal advice.