What moved
The EU AI Act originally applied its rules for high-risk systems listed in Annex III from 2 August 2026. The European Commission proposed a delay on 19 November 2025 through its digital omnibus.
Under the agreement reported by law firms, stand-alone high-risk systems would be covered from 2 December 2027. AI embedded in regulated products would follow on 2 August 2028. The final approach links the start date to the availability of standards and guidance, not to a flat one-year delay.
Sources differ on the exact adoption status at the time of writing. A July 2026 law firm update describes the omnibus as in force. Check the Official Journal text before you rely on a date.
What did not move
The Article 50 transparency obligations are largely unaffected by the omnibus. Providers must also register a system in the EU database for high-risk systems, even if they believe it is exempt from high-risk classification.
A later deadline does not reduce the work. It gives you time to do it properly, with the standards that are still being written.
If you are in Southeast Asia
You may think this is a European problem. It becomes yours when you sell to a European customer, when a customer of yours does, or when a tender asks how you manage AI risk. The questions are similar to the ones Malaysia's draft bill is expected to ask.
Start with the same inventory. Which systems do you run, which of them could touch a decision about a person, and who owns each one. If the answer to the last question is nobody, that is the first finding.