The EU didn't change its mind about the rules. The delay happened because the system needed to enforce them wasn't ready.
To show compliance, companies are meant to follow detailed technical standards that turn the Act's requirements into concrete steps. Those standards are being written by the European standards bodies CEN and CENELEC, which missed their 2025 target and now expect to deliver them in late 2026 at the earliest. Several member states were also behind on appointing the national authorities and assessment bodies that oversee compliance.
Without the standards, companies would have had legal obligations with no clear way to prove they'd met them. So the EU's Digital Omnibus on AI, in force since July 27, 2026, moved the date.
Three things make this different from a simple extension.
The requirements didn't change. Risk management, data governance, technical documentation, logging, human oversight, and controls for accuracy and security are all still required. Only the timing moved.
The new date is fixed. It applies whether or not the standards are finished on time. If they slip again, the deadline doesn't slip with them.
The window is tighter than it looks. If the standards arrive in late 2026, companies will have about a year to put them into practice before December 2027. Organizations that use the months before then to lay the groundwork will spend that year refining. Those that wait will spend it scrambling.
The delay gave the EU time to finish its side of the work. It's an opening for businesses to do the same.
The Omnibus, formally Regulation (EU) 2026/1744, changed two dates:
For most enterprises, the first date is the one that matters. If AI screens job applicants, assesses creditworthiness, supports certain life and health insurance decisions, or determines who gets access to a service, it likely falls under Annex III.
Other parts of the Act are already in effect:
A company with an EU-facing chatbot or generative AI feature may already have obligations today. Legal counsel should confirm which provisions apply to which systems.
The AI Act can apply to companies that place AI systems on the EU market, or whose AI outputs are used in the EU, wherever the company is based. A US employer screening candidates for roles in Germany, or a lender scoring applicants in France, could be in scope.
Most large enterprises would also rather run one AI governance program than one per jurisdiction. Because the EU's requirements are among the most detailed anywhere, building to them tends to cover much of what other frameworks ask for, including the NIST AI Risk Management Framework many US organizations already use.
Much of the work doesn't depend on the standards at all and can start now.
Where to begin:
Build toward, then finalize once the standards land:
Start with a handful of the highest-impact use cases rather than everything at once. The method you build on those carries over to the rest.
Stand-alone high-risk AI systems listed in Annex III have until December 2, 2027. High-risk AI built into products covered by EU product safety law (Annex I) has until August 2, 2028. Both dates were set by the Digital Omnibus on AI, in force since July 27, 2026.
Why was the EU AI Act delayed?
The technical standards companies use to show compliance weren't finished, and several member states were behind on appointing the authorities and assessment bodies that oversee it. The rules themselves did not change.
Did the AI Act's requirements change?
No. Risk management, data governance, technical documentation, logging, human oversight, and controls for accuracy and security are all still required. Only the timing moved.
Is any part of the EU AI Act already in effect?
Yes. Bans on prohibited AI practices and the AI literacy duty have applied since February 2025, obligations for general-purpose AI model providers since August 2025, and most transparency obligations since August 2, 2026.
What should companies do before December 2027?
Start with work that doesn't depend on the standards: inventory your AI use, sort it by risk, assign accountable owners, and train the people using these systems. Then map controls and build evidence, and finalize both once the standards are published.
itD works on both sides of the gap between AI policy and day-to-day practice. On the governance side, we help organizations inventory their AI use, tier it by risk, map controls to frameworks like the EU AI Act and the NIST AI RMF, and set up the reporting that leaders and auditors expect. On the delivery side, our data, security, and change management teams make sure those controls work inside the systems and workflows people actually use. The result is a program that's documented and audit-ready, and that people follow.
The EU used its extra time to get ready. Use yours the same way.
Book a free strategy session to talk through your AI governance roadmap, or take our 5-minute AI readiness assessment to find out where you stand.
This post is for general information and isn't legal advice.
You may also like