"The AI Act got pushed to 2027 anyway." I have heard that sentence in every coaching session since the omnibus vote. It is three-quarters wrong, and it is getting expensive for the people who believe it: while they wait, the August 2, 2026 deadline keeps coming and their competitors are documenting compliance.
Two camps are wrong at the same time. The ones headlining that everything is postponed and closing the file. And the fear merchants waving 35 million euro fines without mentioning that the training obligation was softened. Here is the exact state of the law, sourced, as of July 14, 2026.
What the omnibus actually changed
The digital omnibus was adopted in June 2026: the European Parliament voted on June 16, the Council gave its final green light on June 29. It postpones the high-risk obligations of Annex III (biometrics, recruitment, education, critical infrastructure) to December 2, 2027, and to August 2, 2028 for AI embedded in regulated products. That is where the "AI Act delayed" headlines come from.
It also rewrote Article 4 on training: the duty to guarantee a sufficient level of AI literacy becomes a duty to take measures to develop that literacy, with no guaranteed level. Softened, not removed, and it has applied since February 2025 to providers and to companies that use AI.
What did not move by a single day
August 2, 2026 remains the general application date. On that day the Article 50 transparency obligations kick in (AI-generated content, chatbots), along with their sanctions regime of up to 15 million euros or 3 percent of worldwide turnover, and the national supervisory authorities. Prohibited practices have been in force since February 2025, with fines up to 35 million euros or 7 percent.
In other words: if your company uses AI daily without deploying a high-risk system, your calendar has not changed at all.
What this means in practice
Three files stay on your desk. Usage rules and staff training, because Article 4 still applies and untreated shadow AI is exactly what an audit or a tender will surface. Transparency, if your teams publish AI-generated content or expose chatbots to your customers. And proof: documented measures, program, attendance records, because a duty of means is judged on what you can show.
For executives comparing training options to close that gap, we mapped the market in how much executive AI training costs in 2026, and our executive AI training programs are built to produce exactly that documented proof.
Want to know where your company stands before August 2? A 20-minute call is enough to take stock.
Get in touch →Frequently asked questions
Is the EU AI Act postponed to December 2027?
Not as a whole. The digital omnibus adopted in June 2026 only postpones the high-risk obligations: to December 2, 2027 for standalone Annex III systems, and to August 2, 2028 for AI embedded in regulated products. The general application date of August 2, 2026 stands.
What still applies on August 2, 2026?
Most of the regulation: the Article 50 transparency obligations (with fines up to 15 million euros or 3 percent of worldwide turnover), the new regime against nudification content, and the activation of national supervisory authorities. Prohibited practices, with fines up to 35 million euros or 7 percent, have been in force since February 2025.
Was the AI training obligation removed?
No, softened. Article 4 has applied since February 2025 and required a sufficient level of AI literacy. The omnibus replaces that outcome-based duty with a duty of means: take measures to develop your staff's AI literacy, without guaranteeing a specific level. The obligation still covers both providers and companies that use AI.
Can my company wait until 2027 to deal with this?
Only if you deploy Annex III high-risk systems, and only for those specific obligations. For everything else (transparency, training, prohibited practices) the calendar has not moved. And the fastest risk is not the fine: large clients already ask for compliance evidence in their tenders.
