The European Council adopted its general approach on a package of EU AI Act amendments on March 13, 2026. Every headline led with the deadline extension. That framing is accurate but incomplete.
Three tracks run through the Council’s package. They don’t move on the same schedule. They don’t require the same response. And treating them as a single “deadline delay” story will leave compliance teams underprepared for the provision that matters most right now.
Track 1, The Deadline Extension: What Moved, What Didn’t
The Council’s proposal extends enforcement of high-risk AI system obligations by up to 16 months from the original August 2026 target. Reporting confirms the new enforcement window lands approximately in December 2027, a date the European Commission first proposed in November 2025 to give developers more compliance preparation time.
Two things are not in the extension. First, this is not enacted law. The Council’s position must be reconciled with the European Parliament before the amended text becomes binding. The March 18 committee vote is the next procedural step; a final text requires full Council-Parliament agreement after that. Compliance calendars should carry this date as “proposed, pending Parliament,” not as a confirmed hard deadline.
Second, the extension applies to specific high-risk system provisions. Some August 2026 obligations likely remain active. The Council framed this as a targeted extension. Before treating any August 2026 deadline as cleared, confirm the specific provision against the official Council text. The compliance calendar on this hub will be updated as the official press release becomes available.
Track 2, New Prohibited Practices: The Nudification Ban
This is the element that requires assessment now, before Parliament completes its work.
The Council’s package adds a new category of prohibited AI practices: tools that generate non-consensual sexual or intimate imagery. The Record confirms the Council’s position explicitly prohibits AI practices used to produce such content. The amendment reportedly also addresses child sexual abuse material, per available coverage of the Council package, though that specific provision wasn’t confirmable from the retrieved text at publication. Treat the CSAM element as reported, not confirmed, until the official Council press release is available.
The compliance question for any generative AI provider is not “when does this take effect?” The question is “can my model produce this content?” If the answer is yes, or possibly yes – the product assessment starts now. Here’s why.
Prohibited practices in the EU AI Act framework are not subject to the same phased timelines as high-risk classification requirements. The prohibition framework is designed to be immediate upon enactment. And Parliament is unlikely to remove a prohibition on non-consensual intimate imagery from the final text. The political consensus on this provision is strong regardless of where the deadline extension lands.
Product teams should map their model’s output capabilities against the proposed prohibition language. Legal teams should assess whether existing acceptable use policies and technical controls are sufficient, or whether additional restrictions are needed. The deadline extension gives compliance programs more time on high-risk classification obligations. It doesn’t give generative AI providers more time to assess whether their models can produce prohibited content.
Track 3, Small Company Provisions
The Council’s amendment package reportedly includes provisions addressing small company obligations under the EU AI Act, adjustments to the compliance framework for smaller developers and deployers. The sourcing available for this track is limited. This coverage cannot characterize the specific provisions with confidence, and the official Council press release has not been located at publication. Readers who need precise detail on SME provisions should consult the European Council’s official documentation directly.
This gap is flagged for follow-up coverage once the official text is available.
What Happens Next: The Parliament Reconciliation Process
The Council’s general approach is not the final word. It’s the Council’s negotiating position. What comes next is a trilogue process, informal negotiations between the Council, the European Parliament, and the European Commission, to agree on final amendment text. The March 18 committee vote at the Parliament is the next visible milestone. The timeline from committee vote to final text adoption depends on how much the two institutions’ positions diverge.
For compliance teams, this creates a two-speed response requirement. On Track 2, prohibited practices, the directional outcome is clear enough to act now. The prohibition on AI-generated non-consensual intimate imagery is unlikely to change materially in negotiations. On Track 1 – the deadline extension, the December 2027 date is directionally reliable as a planning horizon, but cannot be treated as a confirmed operative deadline until the final text is adopted.
Uncertainty isn’t a reason to pause compliance planning. It’s a reason to plan in scenarios: what does your program look like if the December 2027 date holds, and what does it look like if Parliament negotiates a different timeline? Build both. Commit to neither until the text is final.