EU AI Act Goes Live
The EU is moving the AI Act from principle to practice: consultations on procedural rules for model access have closed and guidance for general‑purpose AI is being advanced as obligations for high‑risk systems approach in August 2026. That shift turns regulatory discussion into an operational compliance task—documentation, inventories and vendor diligence now have real deadlines. (dig.watch) (circleid.com)
Europe’s Artificial Intelligence Act stopped being a theory exercise when the European Commission closed consultations on how outside researchers can get access to certain general-purpose artificial intelligence models and moved more guidance into final form. What had been a policy debate is now a checklist with dates on it. (ec.europa.eu 1) (ec.europa.eu 2) The calendar is what changed everything. The European Union’s official implementation timeline says rules for general-purpose artificial intelligence models started applying on August 2, 2025, and the main rules for high-risk systems start applying on August 2, 2026. (ec.europa.eu) A general-purpose artificial intelligence model is the base engine that can be reused for many jobs, like one power plant feeding many buildings. The Commission’s guidance says providers of those models must prepare technical documentation, adopt a copyright policy, and publish a summary of training content. (ec.europa.eu) The European Commission also received the final General-Purpose Artificial Intelligence Code of Practice on September 24, 2025 after a drafting process involving 13 independent experts and more than 1,000 stakeholders. The code is voluntary, but Brussels explicitly presents it as a tool companies can use to show they are meeting the law. (ec.europa.eu 1) (ec.europa.eu 2) High-risk systems are a different layer of the law. These are artificial intelligence systems used in areas like employment, education, essential services, and law enforcement, where the Act requires risk management, data governance, logging, human oversight, accuracy, robustness, and cybersecurity. (eur-lex.europa.eu) For companies, that turns compliance into inventory work before it turns into legal work. A bank, hospital, school, or hiring platform first has to know which systems fall into the high-risk bucket, who built them, what data they use, and which contracts push obligations onto a vendor. (eur-lex.europa.eu) (ec.europa.eu) The deployer rules land at the same August 2, 2026 milestone. Article 26 of the Act says organizations using high-risk systems must follow instructions for use, ensure human oversight, monitor operation, and keep logs when those logs are under their control. (artificialintelligenceact.eu) (eur-lex.europa.eu) The model side is already under a clock as well. The Commission says the general-purpose artificial intelligence rules are enforceable by its Artificial Intelligence Office one year later for new models and two years later for existing models already on the market. (ec.europa.eu) That is why the dry-sounding consultations matter. Rules on researcher access, guidance on model obligations, and code-of-practice language decide what evidence a company will need to keep on file when a regulator asks how a model was trained, documented, tested, and monitored. (ec.europa.eu 1) (ec.europa.eu 2) The European Union wrote the law in 2024, but 2026 is when ordinary organizations start feeling it in procurement meetings and product reviews. If a company sells or uses artificial intelligence in Europe, the next few months are less about speeches and more about spreadsheets, audit trails, and vendor calls. (ec.europa.eu) (ec.europa.eu)
