The August transparency deadline absorbed so much attention that the date sitting three months behind it has gone largely unremarked. On 2 December 2026, the grace period attached to the machine-readable marking obligation in Article 50(2) comes to an end for AI systems that were already on the market before 2 August. For anyone who concluded in August that the transparency file was closed, it is not quite closed.
The reason this one slipped past is understandable. When the Digital Omnibus was being argued over, the headline was the deferral of the high-risk regime to 2027 and 2028, which is a much bigger change affecting a much louder constituency. The four-month accommodation on marking was a footnote by comparison, and footnotes with dates in them have a way of arriving anyway.
The distinction that matters here is between the two things Article 50 asks for, which are frequently discussed as one. The human-facing duties, telling a person they are dealing with an AI system and disclosing AI-assisted content where required, began applying in August and were never deferred. Those are about a reader understanding what they are looking at.
Machine-readable marking is a different mechanism aimed at a different problem. It concerns whether content carries technical provenance information that software can detect downstream: markers and metadata travelling with the file rather than a notice a person reads. The two are complementary, and the reason marking got extra time is straightforward: the standards and tooling for doing it interoperably were not mature enough to expect everyone to have implemented them on schedule.
For most professional services firms, this is a question to put to your vendors before it is a question about your own workflow. The marking obligation attaches to the systems producing the content, which means the practical exposure for a deployer runs through the tools in use rather than through anything the firm builds itself. Whether and how it applies to any particular organization depends on what the systems do and how the output is used, which is a question for counsel rather than for a blog post.
The firms most directly affected are the ones generating synthetic media at scale, where provenance markers on images, audio, and video are the clearest case. Teams whose AI use is mostly analysis and drafting sit further from the center of it, though the general direction, that AI-assisted output should be traceable to its origin, applies well beyond the specific technical requirement.
There is time to handle this without drama, which is the whole point of raising it in September rather than in late November. The useful move is to ask each AI vendor in your stack a direct question: what is their position on Article 50(2), what will their output carry by December, and what, if anything, do they expect from you. A supplier who has thought about it will answer plainly. A supplier who has not will tell you about their roadmap, and that is itself a useful answer.
It is also worth noticing the pattern rather than just the date. This is the second time in six months that a deadline moved, and the second time the movement applied to one component while leaving the underlying expectation untouched. Reading the regulation as a series of dates to be survived produces exactly this experience: repeated surprise. Reading it as a direction, that AI-assisted work should be attributable, reviewable, and traceable, makes each individual date considerably less eventful.
Qonera is the AI governance platform for professional teams, built around a structured review and approval workflow where every inference is logged with its provenance and every output traces back to the evidence and the named person who approved it. How we approach our own obligations is set out in the conformity assessment, and the article-by-article mapping is on the EU AI Act page. December 2 is a smaller date than August 2 was, but it is a real one, and three months is enough time to find out whether it reaches you.
This article is for general information only and does not provide legal advice. Regulatory timelines for the EU AI Act are subject to change, and the status described here reflects the position as understood at the time of writing. Organisations should consult qualified legal counsel about how Article 50 and the EU AI Act apply to their specific systems, workflows, and obligations.
Multi-model stress testing, Conflict Heatmap, tamper-evident audit trail, and structured sign-off, built for teams who need defensible AI output.