On 7 May 2026 the Council of the EU and the European Parliament reached a provisional agreement on what is being called the AI Omnibus. Coreper confirmed the compromise text on 13 May, and Parliament’s committees approved the trilogue text on 2 June. The file still needs Parliament’s formal first-reading position, with Council adoption after that, so the new dates are not binding law yet. But the direction is now clear enough for AI builders to adjust their roadmaps.
The headline: high-risk AI obligations under the AI Act are being pushed out, SME and small mid-cap firms get genuine compliance relief, and the documentation burden is being trimmed. If your team is building AI in the EU, your compliance calendar just shifted.
This is not a rollback. The substantive obligations of the AI Act remain. What changed is the timing, the documentation overhead, and who qualifies for simplified compliance.
The new high-risk deadlines
The deal defers application of the most demanding obligations:
- Annex III high-risk systems (the use-case-based category, covering things like AI in credit scoring, hiring, education, critical infrastructure, and law enforcement) now apply from 2 December 2027.
- Annex I high-risk systems (AI as a safety component of a regulated product, or as a regulated product itself) now apply from 2 August 2028.
That is roughly 12 to 16 months of additional runway compared with the original calendar, depending on which category your system falls into.
The delay is not the whole package. The compromise also keeps nearer-term pieces moving: the nudifier and CSAM-related ban, a watermarking deadline on 2 December 2026, an AI regulatory sandbox deadline on 2 August 2027, and clearer product-safety overlap rules. Treat the high-risk delay as extra runway, not as a general AI Act holiday.
Relief for SMEs and small mid-caps
The Omnibus extends simplified compliance to a newly defined category of small mid-cap companies. The Commission’s small mid-cap threshold is firms with fewer than 750 employees, plus either annual turnover of up to €150M or a balance sheet total of up to €129M. That captures a much larger share of the EU AI ecosystem than the original SME definition.
For these firms, the deal trims documentation requirements, lowers fine caps, and broadens sandbox access. The intent is to keep smaller builders moving while the larger players carry the heavier compliance lift.
Centralised enforcement at the AI Office
Enforcement coordination shifts further to the European AI Office. The practical effect is a more consistent interpretation across member states. The fragmentation risk that worried builders during the original AI Act debate just got smaller.
What this means for your team
If you are building or deploying AI in the EU, three concrete actions:
- Re-check your classification. The new high-risk classification guidelines from the Commission landed days later. If your system was on the boundary, the guidance plus the new timeline may change your plan.
- Recalibrate your compliance roadmap. The delay buys time, but the obligations are real. Use the extra runway to build properly, not to defer the work indefinitely.
- Check the SMC threshold. If your firm has fewer than 750 people and is within either the €150M turnover or €129M balance-sheet threshold, you may qualify for simplified compliance. That changes the cost-benefit math on every AI feature you are weighing.
The Omnibus still needs formal adoption. The substance of what was agreed is unlikely to change materially, but watch for the published final text before treating the dates as binding law.
Read the Council’s press release: Council and Parliament agree to simplify and streamline AI rules
Track the Parliament procedure file: 2025/0359(COD)
Read the Commission factsheet on small mid-caps
Share this article
