The EU AI Act Deadline That Didn't Move (And Already Passed)

The EU AI Act timeline is no longer moving as one block: Article 50 transparency obligations took effect on August 2, 2026, while certain high-risk AI requirements were pushed to December 2027, making precise deadline tracking essential for compliance planning.

The EU's Digital Omnibus package has genuinely reshaped the AI Act's timeline — just not evenly, and the part that didn't move is the one I keep seeing people get wrong.

Here's what actually shifted: high-risk AI systems under Annex III — recruitment, credit scoring, education, law enforcement, essential services - now have until December 2, 2027, to comply, a 16-month delay from the original August 2026 date. Product-embedded high-risk systems under Annex I moved from August 2027 out to August 2028. That's real breathing room, and if your compliance plan is built around high-risk conformity assessments, it's fair to feel some relief.

Here's what didn't move: Article 50's transparency obligations, disclosing to people that they're interacting with an AI system, labelling synthetic media and deepfakes, flagging AI-generated content on matters of public interest, took effect on August 2, 2026. That date has already come and gone. The Digital Omnibus never touched it.

There's one nuance worth knowing, and it's narrow. If you already had a generative AI system on the market before August 2, you get until December 2, 2026, to comply specifically with the synthetic-content marking and detection piece of Article 50. That's it; it's the only part of Article 50 with any grace period at all, and it only applies to systems that existed before the deadline. Everything else under Article 50, and any system launched after August 2, was expected to comply immediately, with no transition window.

Separately and easy to lose in the noise about deadlines; new prohibitions on things like AI-generated non-consensual intimate imagery and CSAM took effect December 2, 2026, carrying penalties up to 7% of global annual turnover. That's not a compliance deadline in the usual sense; it's an outright ban, and it wasn't delayed either.

So if I had to correct one thing I hear constantly, it's this: "the AI Act got delayed" is the wrong takeaway. What actually happened is narrower and easy to miss if you only skimmed the Digital Omnibus headlines. The most demanding conformity work, tied to high-risk systems, got real breathing room. The disclosure baseline that applies to nearly everyone using AI-facing tools didn't move at all, and by now it isn't a future deadline; it's a requirement you're either already meeting or already behind on.

If your AI Act compliance plan currently says "we have until 2027," it's worth checking exactly which obligations that timeline actually covers because for a lot of organizations, part of what they think is still ahead of them already happened back in August.