EU AI Act Enters Enforcement Phase With New Labeling Rules

A rulebook moves from paper to practice

European Union regulators began a new phase of artificial-intelligence oversight on August 2, activating transparency duties for chatbots, synthetic media and deepfakes. Providers must tell people when they are interacting with AI, while certain generated or manipulated text, images, audio and video must carry detectable markers or clear disclosures.techreviewafrica +1 The European Commission’s AI Office can also investigate general-purpose AI providers, demand corrective action and impose penalties for breaches of obligations that began applying last year.eualive

The shift gives the bloc’s landmark risk-based law a more visible role in everyday digital services. It also extends beyond model developers: businesses and public bodies that deploy covered systems can bear responsibility for disclosures, including when they publish deepfakes or AI-written material on matters of public interest.reedsmith

Labels arrive with exceptions and grace periods

Article 50 covers systems that interact directly with people, generate synthetic content, perform emotion recognition or biometric categorisation, and create deepfakes or public-interest text.reedsmith AI-generated public-interest writing can be exempt when it has undergone human review and an identifiable person or organisation accepts editorial responsibility. Artistic, satirical and fictional works can use less intrusive disclosure, and personal, non-professional use is generally outside the transparency regime.theguardian

The timetable is not uniform. Existing generative systems have until December 2 to add machine-readable marking, and content produced and published before August 2 does not need retroactive labels.reedsmith Broader high-risk rules for areas including employment, education and critical infrastructure were postponed to December 2027 under the bloc’s recent AI Omnibus changes, meaning the latest milestone does not make the entire law applicable at once.reedsmith

Enforcement will test whether disclosure changes behavior

Violations of the transparency requirements can draw fines of up to €15 million or 3% of worldwide annual turnover.theguardian More than 180 organisations have joined a voluntary transparency code, while Google and Meta are among the companies that have signed it.techreviewafrica +1 The code offers a route to demonstrate compliance but does not replace the legal duties.

Industry groups support identifying deceptive synthetic media but warn that overly broad labelling could produce a new version of cookie-banner fatigue. The Computer and Communications Industry Association has argued that treating routine commercial alterations like deceptive political fabrications may cause users to tune out notices.theguardian Regulators now face the harder test: making labels consistent and useful while proving that formal enforcement can keep pace with rapidly evolving AI products.