In four weeks, on 2 August, the EU AI Act reaches its next milestone. If you followed the news this spring, you know the picture got complicated: the EU institutions agreed to postpone the most demanding high-risk obligations, partly into 2027 and 2028, while other parts still arrive on schedule.
I have sat in enough leadership conversations these past months to know how this lands. The dominant reaction is relief. The delay is read as a pause, and the pause is read as permission to deprioritise. I think both readings are mistakes.
Here is where the shuffle actually landed. On 16 June 2026 the European Parliament approved amendments that push most high-risk obligations to December 2027 and August 2028[A]. What did not move: the Article 50 transparency duties: chatbot disclosure, AI-content marking, deepfake labelling, and the Commission’s GPAI enforcement powers, both live on 2 August 2026[L]. And if the omnibus is not formally adopted in time, the original timeline applies as written[L]. Planning your readiness around the delay is a bet, not a strategy.
Update, 19 July 2026: the Omnibus has since been formally adopted: Parliament on 16 June, Council on 29 June. The postponed dates below become final with publication in the Official Journal. What the finality should change in your planning: read the follow-up.
What changed, and what did not
What changed is timing on specific obligations. What did not change is the direction of travel, the transparency rules that still take effect in August, and the question that was never a compliance question in the first place: can you say, precisely, where AI touches your decisions, your data and your customers?
Most mid-market companies I speak with cannot answer that today. Not out of carelessness. AI arrived tool by tool, team by team, and nobody kept the map. A deadline delay does not fix an inventory problem. It moves the day you discover it under pressure.
What this means for you
If you run a mid-market company, my advice is simple. Use the extra time deliberately. Know your AI inventory: which systems, whose data, which decisions. Classify honestly: most of what you run is probably not high-risk, and knowing that reduces fear as much as risk. And give AI decisions one named owner, because the accountability question arrives with the first incident, not the first audit.
Do not manage AI to Brussels deadlines. Manage it to your own decision quality. Handle the second, and the first takes care of itself.
The timeline, in one view
| Date | What applies |
|---|---|
| 1 August 2024 | AI Act entered into force |
| 2 August 2026 | Transparency obligations (Article 50) apply |
| 2 December 2026 | Watermarking for generative systems placed on the market before August 2026 |
| 2 December 2027 | High-risk obligations, stand-alone Annex III systems (per the provisional Omnibus agreement) |
| 2 August 2028 | High-risk obligations, AI embedded in regulated products, Annex I (per the provisional Omnibus agreement) |
Dates for the postponed obligations become final once the Omnibus is formally adopted and published. The direction does not change either way.
Sources
European Commission: AI Act implementation timeline
Gibson Dunn: EU AI Act Omnibus Agreement, postponed high-risk deadlines
Governance is the real deadline
Here is the number that should worry you more than any date in Brussels: Deloitte finds that while close to three quarters of companies plan to deploy agentic AI within two years, only 21 percent have a mature model for governing autonomous agents[DE]. The AI Act simply forces you to build muscles you will need anyway, an inventory of your systems, clear accountability, documented decisions. Organisations that treat the deferral as a reprieve will meet the same requirements later, in a hurry, at a worse price.
What I see in the field
In a DAX headquarters environment and in mid-sized companies alike, the deadline news produced the same reflex: relief, then a quiet shelving of the compliance workstream. That is the wrong lesson. The teams that used the extra time to build their AI inventory and literacy programme now make faster deployment decisions than the ones waiting for legal certainty that will not come.
We track the AI Act’s moving parts so our clients do not have to re-read Brussels every week: what applies now, what is deferred, and what that means for your deployment decisions. If your obligations are unclear, that is a one-workshop fix.
Personalised answer within 48 hours.
Sources
- EU AI Act Implementation Timeline, artificialintelligenceact.eu; European Parliament amendments of 16 June 2026. artificialintelligenceact.eu
- Latham & Watkins, AI Act Update: EU Resolves to Change Rules and Extend Deadlines, June 2026. lw.com
- Deloitte, The State of AI in the Enterprise 2026 (survey of 3,235 leaders). deloitte.com
The AI Act Just Moved. Your Deadline Did Not.