We Dropped the EU AI Act Enforcement Forecast to 35% — Here's the Reasoning Chain, and the Residual Risk We're Still Watching
TexTak's forecast that 'EU AI Act high-risk enforcement deadline holds at August 2026' sits at 35%, and that number has been declining as the Digital Omnibus process has moved faster than we initially modeled. Today's reporting that trilogue negotiations are targeting a political agreement as early as April 28 — with both Parliament and Council broadly aligned on pushing high-risk obligations from August 2, 2026 to December 2027 — is the most direct evidence yet that the deadline shift will clear before it becomes binding. We want to explain exactly what drove us to 35%, why we haven't moved further, and what residual risk we're still carrying.
The reasoning chain that got us to 35%: three months ago, our main FOR case was procedural — the Digital Omnibus must pass both Parliament and Council, and with only eight of 27 member states having designated competent authorities and harmonized technical standards from CEN/CENELEC still missing, a deadline slip in the legislative process could leave the original August 2 date binding by default. That's still technically true. But the political consensus signal has gotten so strong that procedural risk is now doing the heavy lifting on whatever probability remains above zero. A 101-9 committee vote, the Commission itself proposing the delay, Council agreeing its own mandate on March 13 — that's not a close legislative fight. That's near-consensus with procedural execution as the only remaining failure mode.
Today's reporting adds a precise data point we're weighting heavily: the April 28 trilogue target. If political agreement is reached this week, the remaining probability of the August deadline holding drops sharply. Legislative drafting and formal publication take time, but with genuine trilogue alignment, the instrument gets made. There's also an important nuance in today's coverage that our forecast needs to track carefully: transparency and watermarking requirements for AI-generated content retain the original August 2 deadline even under the proposed delay. That means August 2 remains a live enforcement date for a subset of AI Act obligations — just not the high-risk category our forecast specifically targets. We're not moving the forecast because of watermarking obligations; that's a different question.
The honest residual risk case for YES — the 35% — is narrower than it was but not trivial. Legal analysts cited today explicitly warn that if a deal is not struck before August 2, the original high-risk obligations apply as written. The April 28 target is an aspiration, not a signed agreement. EU legislative process has missed self-imposed deadlines before. And even a completed trilogue agreement requires subsequent formal steps. The window between now and August 2 is tight enough that a collapse in negotiations — unlikely given consensus levels, but not impossible — could leave compliance teams with a live obligation. We're weighting this at roughly 10–15% procedural failure probability, with a smaller residual for the possibility that the agreement is reached but not formally published in time.
What would move us below 20%: a confirmed trilogue political agreement on April 28 or within the following two weeks. At that point, the residual risk becomes almost entirely procedural publication timing, and we'd reflect that. What would move us back toward 50%: any signal of substantive disagreement emerging in April 28 negotiations — on scope, on the December 2027 date itself, or on which obligation categories get delayed. We're watching the April 28 outcome closely. If it clears, the August deadline question is effectively resolved.