EU AI Act fight heats up
European lawmakers are debating proposed amendments to the EU AI Act that civil‑society groups say would weaken safeguards for high‑risk systems. (dig.watch) Business groups argue the law should avoid overlapping AI‑specific rules and call for clear, consistent enforcement through the EU AI Office. (blog.workday.com)
The fight over Europe’s artificial intelligence law has shifted from whether to regulate AI to how much of the rulebook to rewrite before the toughest obligations kick in. (consilium.europa.eu) The European Commission opened that rewrite on 19 November 2025 with its “Digital Omnibus on AI,” a package of amendments to the European Union Artificial Intelligence Act and related aviation rules. The Council of the European Union agreed its negotiating position on 13 March 2026, and the European Parliament adopted its first-reading position on 26 March 2026. (europarl.europa.eu) (consilium.europa.eu) (europarl.europa.eu) The original law entered into force on 1 August 2024 and uses a risk ladder: some uses are banned, some must meet transparency rules, and “high-risk” systems face the strictest checks. High-risk examples listed by the Commission include artificial intelligence used in recruitment and medical software, while banned practices include social scoring. (commission.europa.eu) (eur-lex.europa.eu) Those deadlines are now at the center of the dispute. Under the current timetable, bans on prohibited systems started applying on 2 February 2025, and rules for general-purpose artificial intelligence models and governance started applying on 2 August 2025, while high-risk obligations were due later; the Commission’s amendment ties some of those high-risk duties to the availability of standards and other compliance tools. (artificialintelligenceact.eu) (europarl.europa.eu) The Council backed a fixed delay instead of an open-ended one. Its 13 March text would move the application date to 2 December 2027 for stand-alone high-risk systems and 2 August 2028 for high-risk systems embedded in products, while also restoring some obligations the Commission had proposed to drop. (consilium.europa.eu) Civil-society groups say the rewrite cuts into the parts of the law meant to protect people affected by automated decisions. A 15 April 2026 update from Digital Watch said ARTICLE 19 and more than 40 organizations warned that the package could weaken safeguards for high-risk systems in areas including biometric identification and education, and criticized the Commission for moving without an impact assessment or public consultation. (dig.watch) Business groups are pressing the opposite case: keep the law from stacking a second or third compliance layer on top of existing product rules. A 12 March 2026 joint statement published by DIGITALEUROPE said manufacturers in sectors such as healthcare, machinery, energy and automotive face overlapping documentation and conformity checks, and called for a more coherent role for sector regulators and the European Union Artificial Intelligence Office. (digitaleurope.org) The cost argument is a big part of that lobbying. DIGITALEUROPE’s statement cited Commission analysis estimating that a small or medium-sized company developing a high-risk system could face up to €319,000 in initial compliance costs and up to €150,000 a year after that, while saying some studies put first-year costs closer to €600,000 once certification and staffing are included. (digitaleurope.org) The European Union Artificial Intelligence Office sits in the middle of both arguments. The Commission says the office, established in May 2024, is meant to ensure coherent implementation and enforcement of the law, and the omnibus proposal would turn it into a more central enforcement hub for some systems tied to very large online platforms and general-purpose models. (commission.europa.eu) (europarl.europa.eu) Lawmakers have also added fresh restrictions while debating simplification. The Council said its mandate adds a ban on artificial intelligence used to generate non-consensual sexual or intimate content or child sexual abuse material, showing that the rewrite is not only about delay and deregulation. (consilium.europa.eu) The next phase is negotiation between Parliament, Council and Commission over a final text. What began as Europe’s first broad artificial intelligence rulebook is now also a test of how much enforcement, timing and human-rights protection the bloc is willing to trade for faster rollout and lower compliance costs. (europarl.europa.eu) (consilium.europa.eu)
