EU starts enforcing AI Act rules and AI labels on August 2
The European Commission announced on July 31 that its AI Office and national regulators will begin enforcing a new phase of the EU AI Act on August 2. The law itself entered into force in 2024 and has been applying in stages; this deadline activates enforcement powers and transparency requirements that are especially relevant to general-purpose models and generative AI.
The most visible change is disclosure. Chatbots and other interactive systems covered by the rules must tell people they are interacting with AI. Deepfakes must be labelled, while AI-generated or manipulated content must carry machine-readable marks designed to make detection easier. The Commission says these obligations are intended to reduce deception and manipulation. They do not guarantee that every synthetic item will remain identifiable: metadata can disappear when content is copied or edited, so labels and provenance signals still need to travel across platforms.
The AI Office will oversee providers of general-purpose models, including advanced systems that may pose systemic risks. National authorities are responsible for most other AI systems. The Office can demand information, require access for model evaluations, inspect providers, order corrective measures and, if necessary, restrict a model’s public availability. The Commission says the Office now employs more than 125 specialists. New complaint and whistleblower tools give users, companies and insiders channels to report suspected violations.
Penalties give this deadline commercial weight. Prohibited practices can trigger fines of up to €35 million or 7 percent of worldwide annual turnover, whichever is higher. Other breaches, including general-purpose-model obligations, can reach €15 million or 3 percent. Not every part of the AI Act starts now: rules for many high-risk systems arrive later, including December 2027 for categories listed in Annex III and August 2028 for systems built into regulated products.
More than 180 organisations have signed the EU transparency code, which offers a route for putting the rules into practice. OpenAI said on the same day that it is expanding Content Credentials and SynthID provenance measures across media, including audio, and working toward signals for text.
For AI users, the immediate benefit should be clearer disclosure and stronger routes to complain. For makers and businesses, the message is operational: audit chatbot notices, labelling, provenance, documentation, provider contracts and incident processes now. Enforcement turns compliance from a policy promise into a product requirement.