New Transparency Rules for Artificial Intelligence Set to Take Effect Across the EU
A significant regulatory shift for artificial intelligence in the European Union commences in one week. Beginning on August 2, providers of AI systems must comply with new transparency obligations that directly affect how users interact with chatbots and encounter synthetic media. The rules are enshrined in Article 50 of the AI Act, with the final legal underpinning only recently completed. Regulation (EU) 2026/1744 was published in the Official Journal on July 24 and enters into force on July 27, while the European Commission issued its interpretive guidelines on July 20, less than a fortnight before the deadline.
Mandatory Identification for Machines and Deepfakes
Under the new obligations, providers are required to build chatbots and voice assistants so that individuals can recognize they are interacting with a machine. In addition, outputs from generative AI must carry a technical marker in a machine-readable format, a requirement that applies to images, audio, video, and text. Anyone publishing a deepfake, defined as deceptively realistic content featuring real people, is obligated to disclose its artificial nature. The same disclosure principle applies to AI-generated text published to inform the public on matters of public interest, unless a human has reviewed and assumed editorial responsibility for the content.
The framework establishes clear boundaries for these duties. If an AI system only assists with standard editing tasks, such as fixing spelling and grammar, the marking obligation does not apply. Furthermore, content that is obviously fantastical does not meet the legal definition of a deepfake. The Commission offers an illustrative example of this exclusion: depictions of people flying through the air.
Technical Markers and a Phased Transition
The "machine-readable" requirement specified in Article 50(2) concerns watermarks and metadata that are detectable by automated tools, not necessarily a visible icon for human viewers. Recognizing the need for a practical transition, the legislation includes a specific grace period for existing systems. Generative models already placed on the market before August 2 have an additional four months to implement the technical marking, pushing their deadline to December 2, 2026. In practical terms, this covers nearly every widely used service available today.
Alongside the invisible technical markers, a separate voluntary labeling system is being introduced. The Commission has proposed three distinct icons to indicate AI involvement: one for content partly created with AI assistance, one for fully generated material, and one for content that was only partially modified. Their use is part of the Code of Practice on transparency of AI-generated content and is not legally binding. The icons themselves carry no evidentiary weight, and the duties under Article 50 apply regardless of whether they are displayed or not.
National Enforcement and Future Bans
Enforcement mechanisms are already taking shape across member states. In Germany, the Federal Network Agency has been designated as the central market surveillance authority, a move confirmed when the implementing law cleared the Bundesrat on July 10. The agency is establishing a contact point where individuals can report suspected breaches. For failures to comply with the transparency duties, the AI Act authorizes fines reaching 15 million euros or 3 percent of worldwide annual turnover. The seriousness with which the EU will enforce these rules on major technology providers is currently being tested in an ongoing dispute concerning AI assistants on smartphones.
Looking ahead, the regulatory scope will expand significantly. Two new prohibitions are scheduled to take effect from December 2, 2026, targeting AI systems that produce non-consensual intimate imagery or child sexual abuse material. These offenses fall into the highest penalty tier, with potential fines of up to 35 million euros or 7 percent of annual turnover. By contrast, the stricter rules governing high-risk AI applications have been deferred to December 2027 and August 2028.
Sources: digital-strategy.ec.europa.eu, eur-lex.europa.eu, digital-strategy.ec.europa.eu, www.bundesregierung.de