EU AI Act moves to enforcement
The EU is shifting the AI conversation from principles to procedural enforcement, with core obligations for high‑risk systems set to begin applying from August 2026. The European Commission has closed consultations on draft procedural rules and is now working through implementation details for model access and safeguards. That transition means companies supplying or embedding general‑purpose models will face concrete compliance requirements, not just guidance. (dig.watch, dig.watch)
Europe’s artificial intelligence law is no longer sitting in the “get ready” phase. On 8 April 2026, the European Commission closed a consultation on draft procedure rules that spell out how it could inspect general-purpose models and run cases that may end in penalties. (dig.watch) That sounds dry until you read the details. The draft says a Commission access order could require application programming interfaces, source code, model weights, hosting infrastructure, and even the same access levels a provider gives its own employees. (dig.watch) The European Union Artificial Intelligence Act itself is already law. It entered into force on 1 August 2024, banned the first set of prohibited practices from 2 February 2025, and schedules the main wave of rules for 2 August 2026. (digital-strategy.ec.europa.eu) The law works like a safety code with four shelves. Systems judged to create an unacceptable risk are banned, while “high-risk” systems are allowed only if the provider follows a long checklist on testing, documentation, human oversight, and monitoring. (digital-strategy.ec.europa.eu, eur-lex.europa.eu) A general-purpose model is the opposite of a one-job machine. It is a base model that can be reused for many tasks, so one company can build the model and hundreds of other companies can plug it into chatbots, search tools, coding assistants, or hiring software. (digital-strategy.ec.europa.eu) That is why Brussels started with special rules for those base models before the 2026 deadline for many downstream systems. The Commission says obligations for providers of general-purpose models began applying on 2 August 2025, and its own enforcement powers over those providers begin on 2 August 2026. (digital-strategy.ec.europa.eu) The Commission spent 2025 turning broad law into operating instructions. It opened a targeted consultation on 22 April 2025 to answer practical questions like who counts as a provider, what counts as putting a model on the market, and how the European Union’s new Artificial Intelligence Office will supervise compliance. (digital-strategy.ec.europa.eu) Then it published a voluntary General-Purpose Artificial Intelligence Code of Practice on 10 July 2025. The code is meant to show companies one acceptable route for meeting legal duties on transparency, copyright, and safety and security. (digital-strategy.ec.europa.eu) Two weeks later, on 24 July 2025, the Commission published a template for the public summary of training content. That gives providers a common format for disclosing, at a minimum baseline, what kinds of material were used to train a general-purpose model. (digital-strategy.ec.europa.eu) Now the fight is moving from “what does the law mean” to “how exactly will Brussels use its powers.” The April 2026 draft covers independent experts, confidentiality rules, access to case files, and interim measures that could temporarily stop a model from being made available on the market if the Commission sees an urgent risk of serious damage. (dig.watch) For companies outside Europe, the catch is that the law follows the product, not the passport. If a model provider or software company places covered artificial intelligence systems on the European Union market, the August 2025 and August 2026 dates are not European political theater; they are the compliance calendar. (eur-lex.europa.eu, digital-strategy.ec.europa.eu)
