The EU AI Act December 2026 Deadline: Who It Actually Hits

One paragraph of Article 50 was given until 2 December 2026. The rest has applied since August. Most companies are worrying about the wrong one.

There is a date going round, and it is doing damage. 2 December 2026. People have attached a story to it: that this is when the EU AI Act starts asking questions about AI-generated content, and that there is still time.

The story is wrong, and it is wrong in the expensive direction. Article 50 of the EU AI Act has applied since 2 August 2026. December closes a short extension that was handed to one obligation, for one group of companies.

What the EU AI Act December 2026 deadline actually covers

The extension covers Article 50(2), and nothing else. That paragraph is the machine-readable marking duty. A provider of a generative AI system has to mark what the system produces so a machine can tell it was artificially generated: watermarking, metadata, content credentials, whichever technique holds up.

The Digital Omnibus amendments gave a runway to generative systems that were already on the EU market before 2 August 2026. The European Commission states it directly: a limited grace period is envisaged only for AI systems placed on the market before 2 August 2026, and providers of those systems must comply as from 2 December 2026.

Read the condition again, because it is the whole thing. Placed on the market before 2 August 2026. A generative system that launched on or after that date never had a runway. It was supposed to be marking its output from the day it shipped.

Article 50 started in August. December is only when the last piece of it stops being optional for the companies that were already live.

Most of Article 50 did not wait for December

Article 50 has four obligations that matter to an ordinary business. Three of them came into force in August with no extension at all.

Paragraph 1, the chatbot rule. If someone is interacting with an AI system, they have to be told, unless it is obvious to a reasonably observant person.

Paragraph 3, emotion recognition and biometric categorisation. People put in front of it have to be informed that it is running.

Paragraph 4, the deployer duty. If you publish a deepfake you disclose it. If you publish AI-generated text to inform the public on a matter of public interest, you disclose that too.

Paragraph 4 is the one to sit with, because of how it interacts with the grace period. It is a duty on the deployer, and deployers were given no extension whatsoever. So if the tool you use is still inside its marking runway, that runway belongs to the vendor, not to you. Your own disclosure obligation has been live since August regardless of what your vendor has shipped.

Who the EU AI Act December 2026 deadline applies to

Three groups. Most companies reading this are in the third one.

You provide a generative AI system that was on the EU market before 2 August 2026. This is your date. Machine-readable marking on output, in place by 2 December.

You provide one that launched on or after 2 August 2026. Your deadline is behind you. The grace period was written for legacy systems and you were never in it.

You use AI rather than build it. 2 December changes nothing for you, because nothing about your position changes on that day. Your obligations started in August. If you are not sure which bucket you are in, the obligations checker walks it in about a minute.

That third group is where almost all the confusion lives. An agency drafting client copy with ChatGPT is a deployer rather than a provider. Their date was 2 August 2026, and it went past quietly.

What this means if you only use AI

Nothing about the EU AI Act December 2026 deadline gets you more time, so the useful question is what an enforcement conversation or a client audit would actually ask for. In practice it is three things, and none of them are a policy document.

What AI you use, and what for. Who is responsible for each system, by name. And whether a person reviewed the output before it went out, with something to show for it other than a memory.

That last one is where most companies have nothing. The review usually happened. It just left no trace, which under an enforcement regime carrying penalties of up to 15 million euro or 3% of worldwide annual turnover is uncomfortably close to it not having happened. It is also, increasingly, the thing clients ask about directly on an AI vendor questionnaire.

The deadlines after the EU AI Act December 2026 deadline

The EU AI Act December 2026 deadline is the fifth of eight dates in the Act. Three more follow it.

2 August 2027. General purpose AI models that were on the market before 2 August 2025 have to be brought into line with the rules the newer ones already follow.

2 December 2027. High risk systems, standalone. Moved here by the Digital Omnibus from its original date.

2 August 2028. High risk systems embedded in products that are already regulated under other EU law.

We keep all eight on one page, with what each date does and who it lands on, and it updates itself as they pass. If you want the shape of the next two years in one screen, that is the EU AI Act timeline.

What to do before 2 December

If you build a generative system that was live before August, this is an engineering job and you already know it. Marking on output, detectable by a machine, and as hard to strip as the state of the art allows.

If you use AI, the EU AI Act December 2026 deadline belongs to your vendors. Chasing it costs you the time you need for the one that already went past. Your deadline was 2 August 2026, and the gap to close is evidence. Write down every AI system in use with a named owner against it, and start keeping a record when a person signs off on AI-assisted work. Both of those are a morning’s work the first time and minutes a week afterwards.

That is what Art50 is: an inventory of the AI a company uses, and sealed records of human review that anyone can verify without an account and without taking your word for it. The free plan covers an inventory item and twenty records a month, no card.

Art50 records human review of AI-assisted work and keeps an inventory of the AI a company uses. It provides evidence, not legal advice. Article 50 has applied since 2 August 2026; the full text of the article is here and the Commission’s summary of the transparency rules is here.