Article 50 of the EU AI Act (Regulation (EU) 2024/1689) applies from 2 August 2026. The Digital Omnibus on AI deferred the high-risk rules — stand-alone Annex III systems to 2 December 2027, systems embedded in regulated products under Annex I to 2 August 2028 — and left the transparency obligations exactly where they were. One piece moved: the Article 50(2) machine-readable marking duty for systems already on the market, which gets until 2 December 2026.
So the headline everyone read — the EU delayed the AI Act — describes a deferral that skipped the obligation closest to landing. If you ship a chatbot, a generative feature, or anything that produces synthetic images, audio, video or text into the EU, your deadline did not move.
What actually applies on 2 August 2026?
Four transparency obligations apply in full and unchanged from 2 August 2026. They sit in Article 50 and they split across providers and deployers:
- Article 50(1) — AI interaction disclosure. Providers must design systems that interact directly with people so those people are informed they are dealing with an AI system, unless it would be obvious to a reasonably well-informed, observant and circumspect individual.
- Article 50(3) — emotion recognition and biometric categorisation notice. Deployers must inform the people exposed to the system and process their personal data in line with data protection law. Recital 18 confines “emotion” to states such as anger, satisfaction or shame — it excludes physical states like pain and fatigue.
- Article 50(4) — deepfake and public-interest text labelling. Deployers must disclose artificially generated or manipulated image, audio and video content, and AI-generated text published to inform the public on matters of public interest.
- Article 50(5) — how the notice is given. Clear and distinguishable, at the latest at the first interaction or exposure, conforming to accessibility requirements, and accounting for vulnerable people.
Article 50(6) makes these cumulative with the high-risk and general-purpose AI transparency duties rather than an alternative to them. And the exposure is real money: Article 99(4)(g) puts breaches of Article 50 in the tier of up to €15,000,000, or 3% of total worldwide annual turnover for the preceding financial year if the offender is an undertaking, whichever is higher. For SMEs and start-ups, Article 99(6) inverts that — the cap is the percentage or the fixed amount, whichever is lower.
The one thing that moved
Article 50(2) — the provider duty to mark synthetic output in a machine-readable format and make it detectable as artificially generated — now applies from 2 December 2026 for systems placed on the market before 2 August 2026. Systems placed on the market from 2 August 2026 comply on placing. There is no relief for new products.
Here the sources contradict each other, and the contradiction starts at the top. The Council’s press release of 29 June 2026 says the regulation “reduces the grace period for providers to implement transparency solutions for artificially generated content from 6 months to 3 months, with the new deadline set on 2 December 2026.” Both halves of that sentence cannot be right: 2 August to 2 December is four months. Law firm summaries have repeated the three-month figure because they were reading the press release. The date is the operative fact and the date is 2 December 2026 — write that in the plan, not the arithmetic.
There is a second wrinkle worth checking against the final text. The Commission’s FAQ on the Code of Practice on Transparency of AI-Generated Content describes the transitional period as covering AI systems in scope of Article 50(2) and (4) placed on the market before 2 August 2026. The Council press release describes it narrowly, as a grace period for providers implementing marking. The narrow reading is the safer one to plan against.
What the Omnibus did defer
The deferrals are real, they are substantial, and none of them touch Article 50(1), (3), (4) or (5). Taken from the Council’s own record of the adopted regulation:
| Obligation | Original date | New date |
|---|---|---|
| Stand-alone high-risk AI systems (Article 6(2), Annex III) | 2 August 2026 | 2 December 2027 |
| High-risk AI embedded in regulated products (Article 6(1), Annex I) | 2 August 2027 | 2 August 2028 |
| National AI regulatory sandboxes established by competent authorities | 2 August 2026 | 2 August 2027 |
| Article 50(2) marking, systems on the market before 2 August 2026 | 2 August 2026 | 2 December 2026 |
| New Article 5 prohibition on AI-generated NCII and CSAM | — | December 2026 |
That last row is the tell. This package was not a retreat. It added a prohibition on generating non-consensual sexual imagery and child sexual abuse material — the “nudifier” ban — into Article 5, alongside the deferrals. Marilena Raouna, Cyprus’s Deputy Minister for European Affairs, framed the adoption this way: “by banning AI-generated or manipulated sexual deepfakes and AI-generated child sexual abuse material, we are sending a clear message that technological progress must always go hand in hand with the protection of our fundamental values.”
The regulation also clarifies the AI Office’s supervisory competence over AI systems built on general-purpose AI models by the same provider, while leaving national authorities competent in law enforcement, border management, judicial authorities and financial institutions. Products covered by the Machinery Regulation ((EU) 2023/1230) are exempted from the AI Act’s direct applicability, and a new mechanism lets the Commission limit AI Act requirements by implementing act where sectoral law — medical devices, toys, lifts, watercraft — already imposes similar ones.
Is the Omnibus actually law yet?
As at 16 July 2026, no. The European Parliament endorsed the text on 16 June 2026 and the Council gave its final green light on 29 June 2026, closing procedure 2025/0359(COD). But the Council’s own next-steps note says the act “will be published in the EU’s official journal shortly and will enter into force on the third day after this publication.” No publication has been confirmed at the time of writing.
Until entry into force, the original calendar in Article 113 is the law. That is not a technicality — it is the difference between a plan and a hypothesis. The arithmetic is tight: with entry into force three days after publication, the regulation has to reach the Official Journal by 30 July 2026 to be in force before the 2 August 2026 date it is meant to displace. The co-legislators know this, which is why the Council said it treated this part of the package with “utmost priority.”
You can check the position yourself against the adopted text (PE-30-2026-INIT) and the EU Law tracker entry for procedure 2025/359 rather than against a summary. If you are reading this after publication, the dates in the table above are binding; before it, they are a forecast with very good odds.
What this means for your roadmap
An Article 50 programme cannot be re-baselined to 2027. Marking, labelling and first-interaction notice are engineering and design work against a fixed date, and the Omnibus gives you nothing on any of them except the legacy marking window.
The concrete pieces to have in place:
- Inventory by Article 50 paragraph, not by risk tier. Article 50 catches minimal-risk systems. A support chatbot that will never be high-risk is still in scope of 50(1).
- Marking for generative output across every modality — audio, image, video, text — for anything placed on the market from 2 August 2026, and by 2 December 2026 for what is already out there.
- A decision on the Code of Practice. The Code of Practice on Transparency of AI-Generated Content was published on 10 June 2026; the Commission concluded on 8 July 2026 that it adequately covers Articles 50(2), (4) and (5), and the AI Board adopted its adequacy assessment the following day. Signing is voluntary and severable — the provider section and the deployer section can be signed independently. It is also not conclusive evidence of compliance, in the Commission’s own words.
- Watch for the Commission’s Article 50 guidelines, which the Commission has said will be published ahead of 2 August 2026 and which will settle scope questions the Code deliberately leaves alone — including how far the “obvious” exception in Article 50(1) actually reaches.
The reasonable read of the last twelve months: the EU deferred the obligations that depend on infrastructure it hasn’t built — harmonised standards, notified bodies, designated national authorities — and kept the ones that depend only on you.
Frequently asked questions
Did the EU delay the whole AI Act?
No. The Digital Omnibus on AI deferred the high-risk obligations under Article 6(2) and Annex III to 2 December 2027, and those under Article 6(1) and Annex I to 2 August 2028, plus the national sandbox deadline to 2 August 2027. The Article 5 prohibitions (in force since 2 February 2025), the general-purpose AI rules (2 August 2025) and the Article 50 transparency obligations (2 August 2026) were not deferred.
Do I need watermarking by August 2026?
If the system is placed on the EU market from 2 August 2026, yes — Article 50(2) machine-readable marking applies at the point of placing, with no transitional period. If the system was already on the market before that date, the marking duty applies from 2 December 2026.
What if my system shipped in 2025?
It gets the Article 50(2) marking extension to 2 December 2026. It does not get an extension on Article 50(1) interaction disclosure, Article 50(3) emotion recognition notice or Article 50(4) deepfake labelling — those apply from 2 August 2026 regardless of when the system shipped.
Are the prohibitions delayed?
The existing Article 5 prohibitions have applied since 2 February 2025 and were never in scope of the deferral. The Omnibus adds a new prohibition on AI-generated non-consensual intimate imagery and CSAM, which the Council says takes effect in December 2026.
When will the Omnibus be in the Official Journal?
Not confirmed as at 16 July 2026. The Council said on 29 June 2026 that publication would follow “shortly,” with entry into force on the third day after. Working back from 2 August 2026, publication needs to happen by 30 July 2026 for the new dates to displace the old ones in time.
What is the fine for breaching Article 50?
Article 99(4)(g) sets the ceiling at €15,000,000 or 3% of total worldwide annual turnover for the preceding financial year, whichever is higher. For SMEs and start-ups, Article 99(6) reverses the test: the cap is whichever of the two is lower.