In a significant policy shift, European Union governments and Parliament reached agreement on amendments to the AI Act on May 21, 2026. The so‑called “AI omnibus” package aims to reduce regulatory complexity while preserving the law’s risk‑based framework. High‑risk AI systems in areas such as employment, education, and health insurance now face a compliance deadline of December 2, 2027—pushed back from summer 2026—while AI embedded in physical products like medical devices and industrial machinery must comply by August 2028 (euronews.com).
The revised legislation also narrows the scope of what qualifies as “high‑risk.” Systems that merely assist users or optimize performance are no longer automatically subject to the full regime, a change welcomed by manufacturers but criticized by consumer advocates as a potential weakening of protections (euronews.com). Additionally, machinery has been entirely carved out of the AI Act and will now be regulated under sector‑specific rules—a move seen by some as fragmenting the regulatory landscape (euronews.com).
Crucially, the deal introduces a new prohibition: AI tools that generate non‑consensual sexually explicit images—including deepfakes—will be banned, with enforcement beginning December 2, 2026. This closes a gap in existing legislation that failed to address such harmful content (euronews.com).
These amendments reflect a pragmatic recalibration of the EU’s AI governance strategy—balancing innovation and competitiveness with fundamental rights and safety. By extending deadlines and trimming obligations, the EU aims to ease burdens on businesses while retaining core protections. The new deepfake ban underscores the bloc’s commitment to addressing emerging harms in AI content generation.