The Council of the EU gave final approval on June 29 to the AI Act Omnibus (Omnibus VII), a simplification package that grants industry a 16-month reprieve on high-risk compliance while leaving the summer’s transparency deadline exactly where it was. Parliament endorsed the text on June 16; Official Journal publication is expected in July 2026, with entry into force three days after.

The delay is real. Standalone high-risk systems under Annex III, covering employment, education, credit scoring, and law enforcement, now apply from December 2, 2027. High-risk systems embedded in regulated products (medical devices, toys, radio) slip 12 months to August 2, 2028. National AI regulatory sandboxes move to August 2, 2027.

Brussels isn’t calling this a retreat. The Commission’s Omnibus text sequences obligations to the availability of standards and tools, and the plain reason is that European standardisation bodies missed key deliverables. Harmonised standards for high-risk systems are now expected late 2026. You can’t compel conformity with a rulebook that doesn’t exist yet.

What didn’t move is what matters most for the next 31 days. Article 50 transparency obligations, chatbot disclosures and labeling of AI-generated content, still apply from August 2, 2026. GPAI model rules, in force since August 2, 2025, are untouched. The AI Office retains supervisory competence over GPAI-based systems where model and system share a provider, with carveouts for law enforcement, borders, judiciary, and financial institutions.

Two new prohibitions were added rather than deferred: nudifier apps and CSAM-generating systems are banned from December 2, 2026. The watermarking grace period for systems already on the market before August 2, 2026 was tightened, not loosened, to three months from six, expiring the same December 2 date.

The pattern is legible. Obligations that need standards got time. Obligations that need only a disclosure banner didn’t.

Sources

Sources