EU AI Act transparency rules kick in: deepfakes must be labelled

EU AI Act's transparency rules take effect on August 2

New transparency obligations under the European Union's Artificial Intelligence Act take effect today, August 2, 2026. Companies must now ensure that chatbots and other AI systems clearly tell users they are interacting with artificial intelligence, and that AI-generated images, audio, video and text carry a label identifying them as synthetic. The requirements apply to content produced in a professional context; people who use AI purely for personal purposes are not affected, and there are exceptions for artistic, creative, satirical and purely fictional works.

The rules target deepfakes — realistic-looking text, images, video and sound created or altered with AI. An unnamed EU official, quoted by the AFP news agency, said generative AI makes it possible to create and spread false or misleading information on an extraordinary scale, and that the regulations are designed to let citizens tell the difference between real and artificial content so they can keep trusting what they see, hear and read.

The AI Act is being phased in. AI systems that were placed on the market before today have until December 2, 2026 to adapt to the new transparency requirements. Separately, the so-called digital omnibus amending EU regulation took effect on Monday, adjusting some AI Act timelines and cutting administrative burdens. Under that change, obligations for standalone high-risk AI systems — in areas such as biometrics, critical infrastructure, education, employment, and crime detection and investigation — will apply from December 2, 2027. High-risk systems embedded in products covered by sector-specific EU legislation will have until August 2, 2028.

Slovenia has already passed its own law implementing the AI Act, making it one of the first member states to do so. According to the STA news agency, the law identifies the competent authorities and sets up oversight and support mechanisms, which the government says strengthens legal certainty for both users and developers.

What the AI Act's phased deadlines mean for AI providers

What changes for AI providers from today

The immediate effect is that transparency moves from a voluntary practice to a legal requirement for commercial deployments. Any company placing AI systems on the EU market must build in disclosure that users are dealing with a machine, and must label synthetic content. The December 2 grace period for existing systems is significant: it gives providers time to retrofit, but it also creates a split where new systems face obligations immediately. For companies that have already added provenance and labelling features, the transition is comparatively smooth; for those that have not, engineering work and legal review now start on a deadline.

Why the high-risk rules were pushed back

The digital omnibus delays the heaviest part of the AI Act to 2027 and 2028. The stated purpose is to ease administrative burdens and align obligations with sector-specific product rules. That points to a broader EU balancing act: keeping safety standards while giving industry more preparation time and avoiding duplicative compliance. The cost is a longer period of uncertainty for developers of biometric, employment, education and law-enforcement AI, who must wait to see final implementation details.

What the December 2027 and August 2028 dates mean

The new deadlines split high-risk obligations into two groups. Standalone systems in the listed sensitive areas get a December 2, 2027 start. Systems embedded in products already covered by sectoral EU legislation get until August 2, 2028. For affected businesses, the practical question is not simply when the law applies, but how its requirements will interact with the product rules they already follow.

Slovenia's early implementing law

Slovenia's move to adopt national legislation early shows that enforcement is already being built at member-state level while the EU's own deadlines stretch into 2028. The law designates authorities and creates support mechanisms for implementation. For developers and users in Slovenia, that means a named regulator and clearer national guidance sooner than in countries that have not yet legislated.

Where the uncertainty remains

The exemptions for artistic, creative, satirical and fictional works are not yet detailed in the source reporting, and it is unclear how labels will be technically applied across different types of content. The EU official quoted on the scale of AI-generated disinformation was not named. What is clear is that the regime will continue to develop: the omnibus changes took effect this week, and later phases of the AI Act are still to come.

Compliance deadlines AI providers must meet now

• Audit existing AI systems now. Any system placed on the market before August 2, 2026 has until December 2, 2026 to add chatbot disclosure and AI-content labels — map which products and content workflows are affected before that cutoff.
• Treat professional content as in scope. Labelling applies to commercially produced AI-generated images, video, audio and text; personal use and artistic or satirical works are exempt, but routine marketing and customer-facing output is not.
• Plan high-risk compliance in two waves. Standalone AI in biometrics, critical infrastructure, education, employment, and crime detection must meet the rules by December 2, 2027; AI embedded in products covered by sectoral EU legislation has until August 2, 2028.
• Monitor national enforcement. Member states such as Slovenia are already designating AI Act authorities, so providers should track where their systems are used and which national regulators will supervise them.
• Use the omnibus window to document compliance. The digital omnibus reduces some administrative burdens, but the underlying transparency obligations are in force now — labelling infrastructure built in the coming months will carry into the 2027 and 2028 phases.