What happened
On 31 July 2026, the European Commission announced that from 2 August 2026, the AI Office and national market surveillance authorities begin exercising enforcement powers over general-purpose AI (GPAI) model providers under Chapter V of the EU AI Act (Regulation 2024/1689) — including powers to request documentation, conduct evaluations, require risk-mitigation measures, restrict/withdraw models, and impose fines up to 3% of global turnover or €15 million. Simultaneously, Article 50 transparency obligations take legal effect: interactive AI systems (chatbots, AI agents) must disclose their non-human nature, deepfakes must be labelled, and AI-generated content must carry machine-readable marks (fines up to €35 million or 7% of global turnover for violations of prohibited-practice/transparency provisions). The Commission published implementing guidelines on 20 July 2026 and a first list of 180+ organisations that signed the voluntary Code of Practice on Transparency of AI-Generated Content. This is distinct from — and follows on the heels of — the earlier Digital Omnibus postponement of high-risk (Annex III) obligations to December 2027/August 2028.
Why it matters
This is the moment the EU AI Act moves from a paper framework to an actively enforced regime for the most consequential category of AI systems (frontier/GPAI models) and for the broadest population of AI deployers (anyone operating a chatbot, generative tool, or synthetic-content pipeline reaching EU users). Every GPAI provider (OpenAI, Anthropic, Google, Meta, Mistral, etc.) and every deployer of interactive/generative AI systems in the EU market is now subject to binding Commission investigatory and sanctioning authority.
Action needed
GPAI providers must ensure Chapter V technical documentation, copyright policies, and training-data summaries are audit-ready before 2 August 2026 (or by 2 August 2027 for pre-August-2025 models). All deployers of chatbots/interactive AI and generative content tools must implement Article 50 disclosure/labelling by 2 August 2026 (machine-readable marking deadline extended to 2 December 2026 for systems already on market). Consider signing the Code of Practice for a presumption-of-conformity safe harbor.