The Omnibus Is Dead. The August 2 Deadline Isn't. Here's Why We Still Only Give It 35%.

Let's separate the two questions cleanly, because conflating them is exactly the analytical error that would make this forecast unresolvable.

Question one: Does the August 2, 2026 deadline survive as binding law? Given the Omnibus collapse, the answer appears to be yes — at roughly 65%+ confidence. The legal baseline is intact. Absent a new legislative vehicle clearing both Parliament and Council before August 2 (which now looks nearly impossible given procedural timelines), the date holds. We're not forecasting that question, and readers who interpret our 35% as tracking date survival are reading the wrong forecast.

Question two — what [eu-ai-act-enforcement] actually tracks — is whether consequential enforcement follows from that legally binding deadline. This is where 35% lives, and the Omnibus collapse is directionally good news but not nearly sufficient to move us dramatically. Here's why: the EU AI Office's Article 88 enforcement powers against high-risk system operators activate August 2, 2026, leaving roughly five months between activation and December 31. The historical base rate for novel EU enforcement frameworks producing first major actions in their initial months is poor. The DSA took 12+ months despite intense political pressure and a fully staffed enforcement body. The DMA's first formal non-compliance proceedings against Apple launched well over a year after application. Staffing an institution is not the same as producing consequential enforcement, and we've been careful not to treat it as such. The AI Office reaching operational staffing capacity in Q1 2026 is a necessary precondition — it removes one constraint — but it doesn't tell us anything about enforcement willingness, case selection, or the political calculus of moving against a major US lab while Digital Omnibus successor negotiations may still be live.

One regulatory precision point that matters for how you read this forecast: the AI Office's enforcement posture toward general-purpose AI model providers under Chapter V has been operative since August 2, 2025 — that obligation activated 12 months after entry into force. The August 2, 2026 date governs Annex III high-risk system requirements specifically. We flagged in an earlier draft that these were being conflated, and we want to be explicit: [eu-ai-act-enforcement] is scoped to Annex III high-risk system enforcement, not GPAI model provider actions. The [eu-ai-first-fine] forecast — which sits at 30% — covers the GPAI enforcement question separately.

What would move us above 35%? A pre-August 1 formal investigation notice from the AI Office against a named Annex III operator would be the clearest directional signal — we'd revise upward toward 50-55%. Evidence that the Commission is actively coordinating with designated national competent authorities on a first-action strategy would also shift us. What would drop us below 25%? If only 8 of 27 member states have designated competent authorities by July — which is where we are now — and no additional designations follow, the enforcement infrastructure simply isn't in place to produce consequential action at scale within the window. The forecast doesn't yet account for a possible successor legislative mechanism emerging post-Omnibus before August, or for the scenario where the date technically holds but member-state enforcement diverges so dramatically that no action qualifies as 'consequential' in any meaningful sense.