6 Things the Council Just Changed in the EU AI Act
The Council of the EU just signed off on the amendments to the AI Act and yes, the high-risk deadlines are moving. Let’s have a look at what’s actually in there, because there’s more going on than just a delay.
Timing first, since that’s what everyone’s asking about:
- Annex III high-risk systems: pushed to 2 December 2027
- Annex I high-risk systems: pushed to 2 August 2028
- National AI regulatory sandboxes: authorities now have until 2 August 2027 to get these running
- Transparency obligations for AI-generated content: grace period runs out 2 December 2026
A new practice gets added to the banned list. Generating non-consensual sexual or intimate content, or CSAM, is now explicitly prohibited. That covers tools that produce nude images of real people or strip clothing out of existing photos. This one bites from December 2026.
The AI Office’s job just got clearer. Where a provider builds both the general-purpose model and the system on top of it, the AI Office is the one supervising. National authorities still keep their patch though — think law enforcement, border control, courts, and financial services.
Sectoral law and the AI Act don’t have to fight each other anymore. If you’re building something in Annex I that’s already covered by product-specific rules (medical devices, lifts, toys, watercraft, that kind of thing), and those rules already demand AI-specific safeguards similar to the Act’s, the AI Act steps back rather than piling on.
Machinery gets a carve-out. Products under the machinery regulation that used to sit in Annex I as high-risk are no longer directly caught by the AI Act. Instead, the Commission can add health and safety requirements through machinery-specific secondary legislation.
And the Commission has to actually help you comply. For Annex I providers already juggling sectoral law, the Commission is now on the hook to produce guidance that keeps the compliance burden as light as possible.
Net effect: more breathing room on timing, but also a genuine attempt to stop the AI Act and existing product law from overlapping and creating double work. Worth building into your compliance roadmap now rather than waiting for the deadlines to get closer.