The European Union’s Artificial Intelligence Act has crossed its most significant enforcement threshold. From 2 August 2026, the Commission’s AI Office, working with national authorities, gains the power to investigate and sanction providers of general-purpose AI (GPAI) models, while a fresh set of transparency requirements takes direct effect.
The changes were confirmed in a 31 July press release from the European Commission’s digital-strategy directorate. Although the planned “Digital Omnibus” simplification has pushed back obligations for high-risk AI systems, it did not move the Act’s general application date. As a result, three things now apply: enforcement powers over GPAI providers (fines up to €15 million or a percentage of global turnover), the Article 50 transparency duties, and the full sanctioning framework.
In practice, the transparency rules mean certain AI systems must tell users when they are interacting with AI and when content has been generated or altered by it. For developers shipping chatbots, image generators, or agentic tools into the EU market, that is an engineering requirement, not just a compliance checkbox — labels, watermarks, and disclosure flows need to be built into the product.
The practical takeaway for teams building on OpenAI, Anthropic, Google, or open-weight models: if your system’s output touches EU users, the GPAI obligations are live now. Conformity assessments and technical documentation for the hardest categories can take months to assemble, so starting the compliance process sooner rather than later is the safer path.