EU Lawmakers Fail to Reach Agreement on Revised AI Act
Background of the Negotiations
The European Parliament and EU member states convened a 12-hour trilogue on Tuesday to discuss revisions to the bloc’s AI Act. The talks, held under the rotating Cypriot presidency of the EU Council, ended without a deal, leaving the legislation’s toughest enforcement deadlines unchanged.
Key Points of Disagreement
Negotiators could not reconcile differing views on whether industries already covered by existing sectoral safety rules—such as machinery, toys, and medical devices—should be exempted from the new AI requirements. Some member states and lawmakers insisted on keeping these sectors within the AI framework, while the European Parliament pushed for an exemption.
Proposed Deadline Adjustments
Both the Council and Parliament had previously agreed to push back key dates:
- 2 December 2027 for stand-alone high-risk AI systems.
- 2 August 2028 for high-risk AI systems embedded in products.
The rationale for the postponement is the lack of ready-made technical standards, with the Joint Technical Committee 21 of CEN-CENELEC unlikely to finalize them before December 2026.
Stakeholder Positions
Consumer, medical, and academic groups have opposed the exemption, warning that it could weaken the AI Act’s core safeguards. Industry representatives, such as Neil Shah of Counterpoint Research, argue that dual compliance—meeting both physical and digital safety rules—creates a heavy burden, especially for highly regulated sectors.
Implications for CIOs and Companies
Regardless of the political outcome, August 2 remains a hard deadline for high-risk obligations. CIOs are advised to:
- Begin AI inventory and risk assessments now.
- Develop governance and compliance frameworks aligned with the AI Act’s risk categories.
- Prepare for potential fines, reputational damage, and operational constraints if compliance is delayed.
What Happens Next?
Talks are scheduled to resume in May. If no agreement is reached before August 2, the original high-risk obligations will apply automatically, even if standards or enforcement bodies are not fully operational. Other provisions of the AI Act—such as the ban on unacceptable-risk AI (effective February 2025) and the general-purpose AI rules (effective August 2025)—continue on their existing timelines.
Conclusion
The stalemate highlights the tension between regulatory ambition and practical readiness. While the EU strives to create a unified digital rulebook, industry stakeholders remain concerned about the cumulative compliance load. Until a consensus is reached, organizations should treat the upcoming deadlines as immutable and accelerate their AI governance initiatives accordingly.
