The European Parliament and member states have reached an agreement to amend the EU’s comprehensive AI law, postponing some restrictions, simplifying others, and reducing certain compliance burdens for companies. The amendments still require formal adoption by the Council and the European Parliament. According to EU communications, the changes are intended to deliver “safer and simpler rules” for both citizens and businesses.
What the amendments change
-
The package generally streamlines the oversight and enforcement role of the EU AI Office, extends several compliance deadlines for developers, and simplifies some requirements.
-
Obligations for AI systems classified as “high-risk” — including systems used in law enforcement, critical infrastructure, employment, migration, and personal identification — would be pushed from August 2026 to December 2027.
-
Developers would have until August 2027 to implement supervised sandbox environments that isolate new models during testing.
-
Compliance deadlines for certain AI-driven products, such as machinery and toys, would move to August 2028.
-
Deadlines for watermarking AI-generated output and some transparency requirements would shift to around December 2026.
-
The revisions clarify how personal data may be used in training and deployment: whereas existing rules allow certain personal data categories only when “strictly necessary,” the changes permit using personal data to detect and mitigate bias.
-
The amendments also carve out or clarify exemptions for some products. For example, industrial machinery remains primarily regulated by product-safety laws and therefore is less affected by the AI Act.
-
Lighter compliance obligations and administrative burdens would apply in some cases to smaller companies (fewer than 50 employees with either annual worldwide revenue up to €10 million or total assets up to €10 million) and to so-called “small mid-cap” firms (roughly 250–749 employees with either annual worldwide revenue up to €150 million or total assets up to €129 million).
-
One notable strengthening: the amendments ban generation of sexually explicit images of children and the creation of non-consensual nude images of real people.
Background and drivers of the changes
The EU passed the AI Act in 2024; it entered into force that year but included phased implementation of many provisions. From early in the legislative process critics argued the law imposed excessive burdens on companies without clear safety benefits.
-
In 2023, executives from 163 companies signed a letter calling the legislation “bureaucratic.”
-
In 2025, 110 companies asked policymakers to delay the implementation timeline, arguing the rules were “unclear, overlapping and increasingly complex.” Companies such as Siemens and SAP lobbied for revisions, saying the regulations hindered their operations.
-
Two reports influenced policymakers: an April 2024 report by Enrico Letta highlighted fragmentation across the EU’s 27 national markets that hampered scaling of European firms; a September 2024 report described stalling GDP growth as an “existential challenge” and urged measures to close innovation gaps, accelerate decarbonization and reduce dependencies.
-
In early 2025, the European Commission announced an intention to reduce regulatory burdens, simplify rules and boost economic competitiveness.
-
In February 2026, the European Commission withdrew its proposed AI Liability Directive — a separate, controversial proposal that would have set EU-wide standards for lawsuits over AI-related harms.
Reactions and why this matters
Immediate reactions were mixed. The AI industry generally welcomed the additional flexibility and longer timelines, while consumer groups warned the changes could weaken safety protections. Some media framed the amendments as watering down the law to accommodate business interests. The European Consumer Organization said the deal makes the digital environment less safe and creates dangerous loopholes for AI companies.
The AI Act, in both original and revised forms, aims to address potential AI-induced “systemic risks,” a concept borrowed from finance and infrastructure regulation referring to failures that could ripple across sectors or large swathes of the economy. Assessments of such systemic risks remain partly speculative, while overly strict regulation risks stifling innovation. The revisions seek a balance by easing burdens on developers, giving companies more time to understand and comply with requirements, and clearing the way for continued innovation in critical industries such as manufacturing and semiconductors.
Conclusion
The agreed changes relax and postpone several demanding provisions of the AI Act while preserving key prohibitions and clarifications. The amendments must still be formally approved by the Council and the European Parliament, and debates will continue over how best to balance safety, consumer protection and European competitiveness.
