EU publishes final AI transparency rules ahead of Aug. 2 deadline

The European Commission on Monday published final guidelines on AI transparency obligations under the EU AI Act, giving companies less than two weeks to comply with rules requiring deepfake labels and disclosure of AI interactions before they take effect on August 2, 2026.digital-strategy.europa+1

What the Rules Require

The guidelines, published July 20, clarify the scope of Article 50 of the AI Act, which imposes four categories of transparency obligations. AI providers must design systems so users know when they are interacting with AI, such as chatbots or voice assistants. Providers of generative AI must also embed machine-readable marks in synthetic audio, images, video, and text to enable detection of AI-generated content.bratby+2

Deployers face separate duties: they must clearly disclose when content constitutes a deepfake — defined broadly as AI-generated or manipulated material that could falsely appear authentic — and must label AI-generated text published to inform the public on matters of public interest. Emotion recognition and biometric categorization systems must also be disclosed to those exposed to them.europeansting+2

The guidelines note that AI-supported manipulation of minor technical aspects of pre-existing content, such as background editing, lighting adjustments, or color correction, may fall outside the deepfake definition and thus outside labeling requirements. For content that is "evidently artistic, creative, satirical, fictional or analogous," labeling obligations are reduced to disclosures that do not interfere with the enjoyment of the work.artificialintelligenceact+1

Enforcement and Timeline

Violations can attract fines of up to €15 million or 3 percent of worldwide annual turnover, whichever is higher. However, enforcement will rest with national market surveillance authorities in each member state, whose powers also begin on August 2.europeansting+1

AI systems already on the market before August will receive a short reprieve on the watermarking obligation specifically, with those marking and detection duties deferred until December 2, 2026, following a political agreement reached in May between the European Council and Parliament. The guidelines themselves are non-binding but are intended to serve as a reference for regulators and courts assessing compliance.gtlaw+3

A Voluntary Code Fills the Gaps

The guidelines complement a Code of Practice on Transparency of AI-Generated Content, finalized in June 2026 by independent experts. The voluntary code provides a practical framework including the use of a common icon for labeling deepfakes and recommends a multilayered approach combining visible disclosures with invisible techniques such as metadata and watermarking. Companies that adhere to the code can use it to demonstrate compliance with the AI Act, according to the Commission.techpolicy+3