2 December 2026: the EU AI Act's other Article 50 deadline
Almost everything about Article 50 applies from 2 August 2026. There is exactly one exception, and it's easy to get wrong: the machine-readable marking duty in Article 50(2) (the requirement that AI-generated audio, image, video, and text be embedded with detectable provenance data like C2PA Content Credentials) is deferred, but only for generative AI systems that were already placed on the market or in service before 2 August 2026. For those systems, the marking obligation doesn't bite until 2 December 2026, under Regulation (EU) 2026/1744, the July 2026 amending regulation that wrote the deferral into law.
Why this date exists
Retrofitting machine-readable provenance into an AI system that's already shipping (updating model output pipelines, coordinating with a growing web of downstream tools and integrations) is a real engineering lift, more so than switching on a chatbot disclosure banner. The Commission's compromise gives providers of systems already in the market a short runway specifically for that one duty, while holding the line on everything else. It's not a broader delay of Article 50, and treating it as one is the most common misreading floating around compliance commentary since the Digital Omnibus news cycle.
Who this affects
- Covered by the deferral: a generative image/video/audio/text system that was genuinely on the market before 2 August 2026: its marking duty starts 2 December 2026 instead.
- NOT covered, applies 2 August 2026 regardless: any chatbot or AI system that interacts directly with people (Article 50(1)); deepfake and AI-generated-text-for-public-information labeling (Article 50(4)); any generative system launched fresh on or after 2 August 2026; a business merely using a generative tool rather than providing one, where the marking duty sits with the provider, not the deployer, either way.
What changes on the day itself
For providers of pre-August-2026 generative systems, 2 December 2026 is when the machine-readable marking duty starts applying to them the same way it already applies to anything launched after 2 August. There's no new mechanism, no new penalty tier, and Article 50 breaches carry the same EU-wide exposure throughout: up to €15 million or 3% of worldwide annual turnover. Expect a second, smaller wave of compliance attention and search demand as the date approaches, echoing the run-up to August, worth planning content and outreach around if you're building in this space, the way the run-up to 2 August already has.
Common questions
Does the 2 December 2026 date mean Article 50 enforcement starts in December instead of August?
No. Article 50's chatbot disclosure duty (50(1)) and its deepfake/AI-text labeling duty (50(4)) both apply from 2 August 2026, no exceptions. The only thing that moves to 2 December is the machine-readable marking duty (50(2)), and only for generative systems that were already on the market before August.
We're a business that uses AI image/video tools, not a company that builds them, does this date matter to us?
Only indirectly. The marking duty in Article 50(2) sits with the provider of the generative system, not with a business using it to publish content. What matters to you either way is whether the tool you use marks its output (its documentation and a metadata inspection of a sample export will tell you), and whether you're applying visible AI-content labels under Article 50(4), which was never part of this deferral.
How did 'the AI Act is delayed' become such a common misreading of this?
The Digital Omnibus simplification package pushed the high-risk AI regime's deadlines out to 2027-2028, and a lot of coverage compressed that into a general 'AI Act delayed' headline. Article 50 was left alone in that package, with only the narrow 50(2) pre-existing-system carve-out described here. If your compliance plan is waiting on a broader delay, it's waiting on the wrong article.
What counts as a generative system 'already on the market' before 2 August 2026 versus a new one?
The Commission's own detailed criteria for this line weren't fully settled as of this review, a system substantially updated or re-launched close to the date could reasonably fall into either bucket depending on how significant the change was. If you provide a generative AI system and the timing is close, treat 2 August as your target rather than assuming the December date covers you.
Is there a specific fine tier for missing the 2 December marking deadline, separate from the general Article 50 penalties?
No separate tier, a missed 50(2) marking duty falls under the same EU-wide Article 50 penalty framework as any other breach: fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher.