EU AI Act Omnibus Enters Force, Delays High-Risk AI Deadline to December 2027

Regulation (EU) 2026/1744, the AI Act Omnibus, entered force July 27, 2026, granting a 16-month industry reprieve on the EU AI Act’s high-risk provisions: obligations for employment, credit, law enforcement, and critical-infrastructure AI systems move from an August 2, 2026 binding date to December 2027. Transparency requirements (disclosure of AI interactions, deepfake labeling, notice on emotion-recognition and biometric-categorization systems) still take effect on schedule August 2, as do new prohibitions on non-consensual AI-generated intimate images and CSAM (effective December 2026). The Omnibus also expands the European AI Office’s supervisory authority over general-purpose AI models integrated into DSA-regulated platforms — a centralization of EU AI enforcement power reaching Meta, Google, and Apple.

This corrects the prior 2026-07-25–eu-ai-act-full-enforcement-facial-recognition-bans entry, which characterized the high-risk provisions as taking binding effect August 2 — reporting available before the Omnibus’s late-July entry into force. The net picture: the EU’s symbolic AI-governance architecture (facial-recognition bans, transparency mandates) held to its original timeline, while the costliest compliance burden for industry (high-risk system obligations) was delayed 16 months under industry pressure — a partial retreat structurally similar to the pattern already documented in 1970-12-31–clean-air-act-1970-creates-epa-enforcement-authority-adjacent US regulatory-capture cases, where headline prohibitions proceed on schedule while the load-bearing compliance requirements get quietly deferred.