Many plans jump straight from 2 August 2026 to 2 December 2027, because the high-risk obligations moved in that direction. That skips a date. The next hard AI Act date is 2 December 2026, and it brings two things that have nothing to do with high risk.
First, the transitional period for the machine-readable marking under Article 50(2) ends. Second, new prohibitions start to apply on the use of AI for child sexual abuse material and non-consensual intimate content, added by Regulation (EU) 2026/1744.
What ends on 2 December 2026?
Article 50(2) requires providers of AI systems that generate synthetic audio, image, video or text to mark that output in a machine-readable format as artificially generated or manipulated. That duty has applied since 2 August 2026.
For systems already on the market at that point, a transitional period runs until 2 December 2026. It is the only transitional period inside Article 50, and it is narrow: it covers paragraph 2 only, and only those existing systems.
Who is and is not covered
Covered: a generative system you provide that was already on the market before 2 August 2026. Not covered: a system placed on the market after that date, because the marking duty applies to it immediately. Also not covered: the visible disclosure of deepfakes under paragraph 4, because that duty sits with the deployer and has no transitional period.
The marking has to be effective, interoperable, robust and reliable as far as technically feasible, taking into account the state of the art. That is not a licence to do nothing, but it does acknowledge that detection technology is moving. Record which technique you chose and why, because that reasoning is your evidence.
Which prohibitions are added?
Regulation (EU) 2026/1744 extended the list of prohibited practices with the use of AI for child sexual abuse material and for non-consensual intimate content. From 2 December 2026, technical safeguards apply as well.
Prohibited practices are the heaviest category in the AI Act. Unlike transparency or high risk, there is no route to compliance: the practice is simply not allowed. For most organisations this is not an operational change, but it does touch anyone providing or reselling generative image systems, and anyone writing policy on what staff may do with image generation.
What comes after that?
| Date | What |
|---|---|
| 2 December 2026 | End of the transitional period for machine-readable marking (Article 50(2)) and the new prohibitions on deepfakes and non-consensual intimate content |
| 2 August 2027 | AI regulatory sandboxes operational in the member states |
| 2 August 2027 | End of the transitional period for GPAI models already on the market before 2 August 2025 |
| 2 December 2027 | Core obligations for standalone Annex III systems, including the Article 27 FRIA |
| 2 August 2028 | AI embedded in regulated products under Annex I |
The move of Annex III to 2 December 2027 creates room, but it is not a reason to wait. An AI register, role determination and classification take months, and the FRIA builds on all three. Starting in 2027 is starting late.
What do you do between now and December?
Determine which of your systems fall under paragraph 2
Find the systems you provide that generate synthetic output. Establish per system whether it reached the market before or after 2 August 2026, because that decides whether you still have time.
Choose and document a marking technique
Record which technique you use, why it fits the state of the art, and how you establish that the marking actually works. Without that reasoning on paper you may have met the duty without being able to show it.
Update your policy on image generation
The new prohibitions touch what may happen with generative image systems. Check that your AI policy is explicit about it, and that staff know where the line sits.
The full breakdown of Article 50 is in the practical guide, and the current state of every date in the deadline overview. Organisations that want to handle the marking question and the accompanying file in a structured way can turn to Embed AI.
Frequently asked questions about 2 December 2026
Sources
Newsletter
Every Tuesday, the AI Act week ahead in 5 minutes
A practical briefing on deadlines, new guidance and enforcement, so you know what matters this week. No spam and you can unsubscribe in one click.
Practical and short · No spam · One-click unsubscribe