The European Union has moved to soften and delay parts of its landmark AI Act, adjusting several of its most demanding requirements after sustained pressure from businesses and policymakers who warned that the original timeline risked undermining European competitiveness.
What’s new: The European Parliament and EU member states have agreed on a package of amendments to the AI Act that postpone certain high-impact obligations and simplify elements of enforcement. The revised framework still awaits formal approval by both the Council and Parliament, but officials have described the changes as introducing “safer and simpler rules for both citizens and businesses.”
At the core of the revision is a recalibration of timelines for systems classified as high-risk—those used in areas such as law enforcement, critical infrastructure, employment, migration, and identity verification. Compliance deadlines for these systems have been pushed back to December 2027, compared with the previous August 2026 target.
Developers will also be given until August 2027 to implement supervised sandbox environments designed to test AI systems in controlled conditions before deployment. Additional obligations for AI-enabled products, including machinery and toys, have been extended to August 2028, while transparency requirements such as watermarking AI-generated content are now expected to take effect around December 2026.
How it works: The amendments streamline the role of the EU AI Office, narrowing and simplifying oversight responsibilities while extending compliance timelines across multiple regulatory categories. The revisions also adjust how personal data can be used in AI training and deployment. Where previous rules limited certain data usage to cases deemed “strictly necessary,” the updated version allows broader use of personal data for detecting and mitigating bias.
The changes also introduce exemptions and lighter obligations in specific cases. Industrial machinery, already covered by existing product safety laws, is largely carved out of the AI Act’s scope. Smaller firms—defined as those with fewer than 50 employees and limited turnover or assets—along with “small mid-cap” companies, will face reduced administrative burdens.
One of the few areas of tightening remains unchanged: the amendments reinforce bans on generating sexually explicit images of children and non-consensual nude imagery of real individuals.
Behind the news: The AI Act, first adopted in 2024, was designed as the world’s most comprehensive attempt to regulate artificial intelligence. From the outset, however, it faced criticism from industry leaders who argued it imposed heavy compliance costs and uncertain legal obligations.
By 2023, more than 160 companies signed an open letter describing the proposed rules as overly bureaucratic. A year later, additional industry groups called for delays, citing overlapping requirements and unclear enforcement pathways. Major European firms, including Siemens and SAP, also voiced concerns that regulatory complexity was slowing innovation.
Two influential policy reports helped shape the current revisions. One, published in 2024 by former Italian Prime Minister Enrico Letta, highlighted structural fragmentation across EU markets that limits the scalability of European firms compared to the United States and China. Another report in late 2024 framed Europe’s slower growth as a competitiveness challenge requiring regulatory simplification, innovation support, and reduced dependency on external technologies.
In early 2025, the European Commission signaled a broader policy shift toward reducing administrative burden and improving regulatory efficiency. This direction was reinforced in February 2026, when the Commission withdrew its proposed AI Liability Directive, which had aimed to standardize legal accountability for AI-related harms across the bloc.
Public response: Reactions to the amendments have been sharply divided. Industry groups broadly welcomed the added flexibility and extended timelines, while consumer advocacy organizations warned that easing safeguards could weaken protections against harmful AI systems. The European Consumer Organisation argued that the changes risk creating loopholes that reduce overall digital safety.
Why it matters: At its core, the AI Act is built around the idea that advanced AI systems may generate “systemic risks”—failures capable of cascading across industries and affecting large parts of society. While the magnitude of those risks remains debated, policymakers also face pressure to avoid regulatory frameworks that could slow innovation in strategic sectors such as manufacturing, semiconductors, and software.
The revised approach reflects an attempt to rebalance those priorities: maintaining core safety principles while giving companies more time and flexibility to comply, test systems, and scale responsibly within the European market.
We’re thinking: The original framework of the AI Act contained real ambiguity and operational complexity that risked uneven enforcement. The revised version appears to ease some of those pressures without dismantling the structure entirely. Whether it ultimately improves both safety and competitiveness will depend less on the law’s wording and more on how consistently it is implemented across the bloc.
