Europe turns AI rules into compliance work

Europe’s artificial intelligence law is no longer a debate about future guardrails. The European Union’s main deadline for most of the Artificial Intelligence Act is August 2, 2026, so companies are now turning broad legal text into checklists, audit trails, and product paperwork. (ai-act-service-desk.ec.europa.eu) The law does not arrive all at once. The Artificial Intelligence Act entered into force in 2024, some bans and literacy duties started earlier, rules for general-purpose artificial intelligence models start applying on August 2, 2025, and the wider regime applies on August 2, 2026, with full rollout by August 2, 2027. (digital-strategy.ec.europa.eu) (europarl.europa.eu) (ai-act-service-desk.ec.europa.eu) The first companies feeling this most directly are the ones building foundation models. The European Commission says general-purpose artificial intelligence models are the base layer for many downstream systems, so the law puts direct duties on the model provider, not just on the app company using the model. (digital-strategy.ec.europa.eu) To make that workable, Brussels backed a voluntary General-Purpose Artificial Intelligence Code of Practice. The Commission says 13 independent experts drew it up with input from more than 1,000 stakeholders, and it published the final version on July 10, 2025. (digital-strategy.ec.europa.eu 1) (digital-strategy.ec.europa.eu 2) “Voluntary” does not mean “ignore it.” The Commission describes the code as a tool to help providers show they meet the law’s duties on transparency, safety, and documentation, while also saying companies can prove compliance through other methods if they choose. (digital-strategy.ec.europa.eu 1) (digital-strategy.ec.europa.eu 2) That changes the job inside an artificial intelligence company. A lab that used to focus on model training now also needs version records, testing evidence, incident processes, and files that explain what data, safeguards, and limits sit behind a release. (digital-strategy.ec.europa.eu) (raconteur.net) A second European law may land on some of the same firms from a different angle. The Digital Services Act applies its toughest duties to very large online platforms and very large online search engines, which the Commission defines as services with more than 45 million users in the European Union. (digital-strategy.ec.europa.eu) That is why OpenAI suddenly sits in two regulatory conversations at once. Reuters reported on April 10, 2026 that the European Commission was analyzing whether ChatGPT should be treated as a large online platform under the Digital Services Act after OpenAI published user numbers above the 45 million threshold. (kfgo.com) If that designation happens, the overlap is the real story. One law asks how a model is built and documented before it reaches the market, while the other asks how a mass consumer service manages systemic risks once tens of millions of people are using it. (digital-strategy.ec.europa.eu 1) (digital-strategy.ec.europa.eu 2) So the European message in 2026 is not “slow down and wait for guidance.” The guidance is already on the table, the main compliance date is less than four months away, and the companies that sell models into Europe now have to run artificial intelligence like a regulated product line, with lawyers, engineers, policy teams, and auditors all working from the same file. (digital-strategy.ec.europa.eu) (ai-act-service-desk.ec.europa.eu)