The EU AI Act has been in force since August 2024, with tiered compliance obligations activating on a rolling schedule. Obligations for general-purpose AI models became enforceable twelve months after the Act entered into force, August 2, 2025. The next major threshold: high-risk AI system compliance obligations, currently set for August 2026.
That deadline may move. According to reports, a preliminary agreement has been reached at the Council of the European Union level that would restructure the implementation timeline under a proposal referred to as Omnibus VII. The primary source for the specific dates is not independently confirmed in this package, the original reporting URL is inaccessible. TJS is working to verify the primary source. All Omnibus VII-specific claims below use qualified language pending that confirmation.
The reported changes, if confirmed. Stand-alone high-risk AI systems would face a revised compliance deadline of December 2, 2027, a 16-month extension from August 2026. AI systems embedded in regulated products covered under Annex II would face a separate revised deadline of August 2, 2028. Reports also indicate the proposal includes a new explicit prohibition on AI-generated non-consensual intimate imagery and child sexual abuse material, though this addition has not been independently confirmed.
Why it matters. Whether or not the extension is confirmed, the compliance signal is already present. A reported Council-level preliminary agreement on deadline restructuring tells compliance professionals three things. First, implementation at scale is proving harder than the original timeline anticipated. Second, the two-tier structure, stand-alone systems vs. Annex II embedded products, will face different timelines, requiring separate compliance tracks. Third, the addition of a new prohibition layer, if confirmed, means the compliance surface is expanding even as the timeline extends.
Context. The EU AI Act’s existing penalty provisions remain in force regardless of any deadline adjustment. GPAI model obligations, already active since August 2025, are not affected by a high-risk deadline extension. The authoritative EU AI Act reference confirms the current enforcement structure and Chapter V penalty framework. A preliminary Council agreement is not the same as formal adoption, it is a negotiated position that still requires full legislative procedure to become binding.
What to watch. Formal adoption confirmation is the critical trigger. Until a Council regulation is formally adopted and published in the Official Journal of the European Union, August 2026 remains the operative deadline. The prudent compliance posture is to continue building toward August 2026 while monitoring for formal confirmation. Organizations that pause programs now, betting on an extension that isn’t yet law, are taking on unnecessary risk.
TJS synthesis. Treat reported extensions as planning inputs, not permission to pause. The compliance infrastructure you build toward August 2026 will be required by December 2027 at the latest under the reported scenario, and possibly sooner if formal adoption doesn’t follow. The extension, if real, gives you more time to do it right. It doesn’t give you permission to stop.