EU Council Agrees to Delay AI Act High-Risk Rules by Up to 16 Months
The EU Council has backed a proposal to push enforcement of high-risk AI system obligations from August 2026 to as late as December 2027 or August 2028, citing missing standards and political pressure from the United States.
The EU Council agreed its mandate for the Digital Omnibus package on March 13, endorsing a proposal to delay enforcement of high-risk AI system rules by up to 16 months. Stand-alone high-risk AI systems would face compliance deadlines of December 2, 2027. High-risk systems embedded in physical products — medical devices, machinery, vehicles — get until August 2, 2028. The original deadline for both was August 2026.
Why the Delay Is Happening
The official rationale is technical: the European Commission missed its February 2026 deadline for publishing guidance on high-risk AI system requirements. The standards bodies responsible for harmonized technical standards are also behind schedule. Without finalized standards, companies can’t complete conformity assessments. Without conformity assessments, enforcement would create legal uncertainty for compliant companies and require regulators to improvise.
That’s a real problem. But the political context is harder to ignore. U.S. officials have applied sustained pressure on Europe to loosen AI regulation, framing aggressive AI rules as an economic and strategic liability. The Digital Omnibus package, which this mandate feeds into, also includes deregulatory moves across other digital policy areas. The delay isn’t happening in isolation.
What Changes During the Extension
The delay buys time, but the requirements don’t disappear. Companies building high-risk AI systems — hiring tools, credit scoring, critical infrastructure, law enforcement applications, medical diagnostics — still need to build toward full compliance. The extension is conditional: the Commission must confirm that the required standards and conformity infrastructure are actually in place before the new deadlines trigger.
If standards slip further, the dates slip further. That conditionality is new to the EU’s AI enforcement approach and is worth watching. It gives the Commission flexibility to adjust without legislative action, which is either a sensible safety valve or a mechanism for indefinite delay, depending on how it’s used.
Two Additions Worth Noting
The Council added a new prohibition not in the original text: AI systems that generate non-consensual sexual and intimate content, or child sexual abuse material, are now explicitly banned under the AI Act. This was uncontroversial and moved quickly.
MEPs are also pushing to extend the watermarking requirement — mandatory disclosure when audio, images, video, or text is AI-generated — until November 2026. The Commission had proposed extending to February 2027. The shorter extension signals Parliament wants faster implementation on transparency measures even as enforcement on risk classification slows.
The Council mandate now heads to trilogue negotiations with the European Parliament. Final text is unlikely before late 2026.
Source: EU Council Press Release, March 13, 2026 — consilium.europa.eu; IAPP