The number is 89. That’s how many days remain until August 2, 2026, the date the EU AI Act’s compliance obligations become binding for providers and deployers of Annex III high-risk AI systems, as well as for GPAI model providers who have crossed the systemic risk threshold.
There is no extension coming. The April 29 trilogue session on the AI Omnibus Regulation ended without agreement after 12 hours of negotiation. The next scheduled session is May 13. Per the EU AI Act tracker, the August 2 deadline reflects Article 113’s general application provision and remains legally binding regardless of where Omnibus negotiations stand.
The compliance picture is concrete. Per Epoch AI’s compute tracking data, 12 AI models have now crossed the 10^25 FLOP threshold, placing them in presumptive GPAI systemic risk classification under Article 51. These are not marginal cases at the edge of the threshold. They are models already in widespread enterprise deployment. Their providers face adversarial evaluation obligations, cybersecurity incident reporting requirements, and mandatory disclosure of training data summaries, all due August 2.
For Annex III deployers, AI systems used in employment, credit, education, critical infrastructure, and other high-risk categories, the August 2 deadline requires conformity assessments, technical documentation, human oversight measures, and transparency obligations toward affected individuals. None of these are lightweight checkboxes.
Separately, EU AI Act Article 57 requires Member States to establish or designate national AI regulatory sandboxes by the same deadline. This obligation sits with governments, not with AI companies directly, but it shapes the compliance landscape: sandboxes are the mechanism through which deployers can test regulated systems under supervised conditions. Member States that miss the designation deadline leave deployers without that option.
The civil society pressure on the Omnibus negotiations is significant. Reportedly more than 40 organizations, according to reporting from multiple outlets, have lobbied against the Omnibus, arguing it weakens fundamental rights protections included in the original Act text. Their concern is that the Omnibus negotiations could dilute existing requirements rather than merely adjust timelines.
What to watch: the May 13 trilogue. If that session reaches agreement, the Omnibus could modify some August 2 obligations, particularly for smaller providers and deployers. If it fails again, compliance teams should treat August 2 as fixed and proceed accordingly. Waiting for the Omnibus to resolve before beginning conformity assessment work is not a defensible compliance posture with 89 days on the clock.
May is the last month in which operational preparation is feasible without entering crisis mode. Technical documentation for Annex III systems typically requires 60-90 days to complete properly. That window closes at the end of this month if organizations haven’t started. Prior TJS coverage on the deadline addressed the extension question directly, the answer hasn’t changed.
The non-obvious implication: the May 13 trilogue failure scenario isn’t just a compliance planning problem. It’s a market structure signal. If the Omnibus fails, GPAI providers above the FLOP threshold face the most demanding regulatory environment for AI systems that exists anywhere in the world, without the procedural adjustments the Omnibus was designed to provide.