EU's big AI rulebook
The EU AI Act has moved into a decisive phase and will become binding from August 2026, imposing risk tiers and strict compliance for 'high‑risk' and 'general purpose' AI with fines up to 7% of global turnover. Lawmakers also aim to ban certain harmful uses — potentially including AI that generates nonconsensual sexual images — while extending some timelines; the UK has refused a mandatory 'opt‑out' copyright exemption for AI and Canada is watching Europe's experiment closely. (securityscientist.net; en.fasoo.com; insurancejournal.com; computerweekly.com; bullsource.com)
Two European Parliament committees — Civil Liberties (LIBE) and Internal Market (IMCO) — voted this week to adopt an explicit ban on “nudifier” apps that create non‑consensual sexually explicit images as part of the Digital Omnibus text. (rte.ie)) EU member‑state ambassadors, speaking under Cyprus’s rotating presidency, formally backed prohibiting AI practices that generate non‑consensual sexual content and AI‑created child sexual abuse material in a political agreement reached last week. (news24.com)) The European Commission’s Digital Omnibus proposed shifting Annex III “high‑risk” compliance toward the end of 2027 (December 2027) as part of a package to simplify overlapping digital rules. (euractiv.com)) Parliament committees countered the Commission’s timeline by proposing a shorter postponement — recommending a new cut‑off of 2 November 2026 for some deadlines rather than the Commission’s later date — a stance that will feed into upcoming trilogue talks with the Council. (rte.ie)) Obligations for providers of general‑purpose AI models entered into application on 2 August 2025, and the Commission issued guidance in July 2025 to clarify providers’ transparency and risk‑management duties for GPAI systems. (digital-strategy.ec.europa.eu)) National enforcement is already being set up: Finland’s implementing acts for AI supervision entered into force on 1 January 2026, naming Traficom as the single contact point and planning around ten market surveillance authorities to share oversight tasks. (valtioneuvosto.fi)) The UK government has publicly ruled out forcing creators into a mandatory “opt‑out” copyright exemption for AI training in its current policy, a decision that legal and industry commentators say could shape transatlantic approaches to content‑use rules. (computerweekly.com)) Canada has moved to deepen ties with the EU on AI governance — signing a non‑binding Canada‑EU AI MoU on 8 December 2025 and publishing a national AI strategy consultation summary on 2 February 2026 as Ottawa seeks standards alignment while supporting domestic AI firms. (canada.ca))
