The European Union further delays the Annex III high-risk compliance deadline beyond 2 December 2027.

Question

Six frontier AI models put this at 34% by 28 Sept 2029. Full reasoning from every model.

WillHappen.ai · Accepted Answer

34% — the median of 6 frontier models, resolving by 2029-09-28.

Answer

Gemini (Google): 34% — Turns on whether CEN-CENELEC delivers harmonised standards in time for conformity bodies to certify high-risk AI by mid-2027. EU legislation delays like MDR and EUDR show willingness to push dates, but secondary postponements face steep trilogue resistance. A further delay requires a formal Commission proposal and Council-Parliament approval, which rarely clears without severe industry gridlock. Major backlogs in notified body capacity by Q1 2027 would push odds well above 60%.

Answer

Claude (Anthropic): 32% — Turns on whether industry lobbying and implementation gaps force another Commission delay proposal before late 2027. EU already delayed once (2026/1744), showing willingness to push deadlines when compliance infrastructure (standards, notified bodies) lags. Base rate for repeated regulatory delays in complex EU tech regulation is moderate-to-high, especially with AI Act's history of amendments. Counterpoint: EU may want to show firmness after one delay; no active delay proposal is reported yet as of Sept 2026, reducing near-term certainty.

Answer

GPT (OpenAI): 37% — This hinges on whether AI Act implementation failures become politically costly before 2027 compliance begins. The Commission, Parliament and Council would need to enact an amending regulation before 2 December 2027. The 2026 postponement shows capacity concerns can move the deadline, but repeat EU legislative delays within 15 months are uncommon. I would raise this sharply if harmonised standards, notified bodies, or Commission guidance are still materially incomplete in 2027.

Answer

Grok (xAI): 38% — A second AI Act amendment must be proposed and adopted before or after the 2027 date. Commission must table an amending regulation; EP and Council then need 6–12 months to co-legislate. EU already postponed Annex III once (Reg 2026/1744); further slips occur mainly when CEN-CENELEC standards miss 2027. Post-deadline retroactive change is rarer than pre-deadline delay, pulling the odds below 50.

Answer

DeepSeek (DeepSeek): 34% — Turns on whether the Commission's promised simplification package reopens the AI Act's core deadlines before the 2027 date. The Commission's 2025-26 'digital omnibus' simplification push has already reopened the AI Act once, showing appetite to move dates when industry complains. But the December 2027 date was itself the compromise from that first delay, and member states plus Parliament resisted further slippage, so a second extension needs a… Base rate for the EU re-extending a freshly-set AI Act deadline within two years is low; the development that flips me is a formal Commission proposal to amend Article 6…

Answer

Qwen (Alibaba): 31% — EU AI Act has already been delayed once; further delay requires political will to revisit, which is low but nonzero given implementation friction. The EU already granted one postponement via Reg 2026/1744, showing willingness to move deadlines when industry readiness lags, but a second delay signals legislative… Member states and the Commission would need to agree on a new legislative act before Dec 2027; the window is ~14 months, and no formal proposal or political momentum for… The strongest trigger would be a change in Commission leadership or a major industry crisis post-2027 showing systemic non-readiness; absent that, the status quo…

The European Union further delays the Annex III high-risk compliance deadline beyond 2 December 2027.

The U.S. Congress votes on advanced AI supply chain or sustainability bills following the November 2026 recess.

The UK government introduces legislation establishing a dedicated AI oversight body.

OpenAI and Anthropic provide new models to UK testers following completion of the U.S. government review.

Oregon publishes standards for third-party AI safety reviews or a kill-switch mandate for frontier models.