Starting this Sunday, August 2, new mandatory transparency requirements for artificial intelligence-generated content will come into force across the European Union, as the bloc rolls out the first phase of its landmark comprehensive AI regulation. The core mandate of the new rules is simple but transformative: any AI system, from large language model-powered chatbots to AI-generated text, images, audio or video, must clearly disclose its artificial origin to end users, so that Europeans can immediately distinguish between authentic human-created content and synthetic AI output.
Policymakers specifically highlight deepfakes—AI-generated or manipulated text, images, video and audio designed to mimic real content—as a key area of concern. Officials note that generative AI has enabled bad actors to create and spread disinformation at an unprecedented scale, with content tailored to target specific audiences and disseminated in near real time, eroding the public’s ability to tell fact from synthetic content. The new rules are designed to preserve consumer trust in the content they encounter online, by ensuring clear disclosure of AI creation.
Under the framework, companies can meet the requirement through a range of methods, including embedded watermarks and other detectable digital markers that flag AI origin. Non-compliance carries the threat of substantial fines, though the EU has clarified that the rules only apply to professionally generated AI content; individuals using AI tools for personal, non-professional purposes are entirely exempt from the mandate. AI-generated text intended to inform the public on matters of general interest that lacks human editorial oversight is also required to carry an AI label, while work categorized as artistic, creative, satirical or fictional receives a full exemption. All existing AI systems operating in the bloc have until December 2 this year to bring their operations into full compliance with the new requirements.
The policy has drawn its share of criticism, with critics arguing that the rules place overly burdensome compliance demands on businesses, and note that rapid, widespread adoption of AI across all sectors could eventually lead to near-universal labelling requirements. But Ashley Casovan, a representative of the International Association of Privacy Professionals, downplayed these concerns, telling AFP that similar predictions of unmanageable compliance burdens have accompanied past new regulatory requirements, and that businesses will adapt over time.
Many of the world’s largest technology companies have already begun implementing their own AI labelling systems in anticipation of the new rules. TikTok has required creators to label AI-generated images, audio and video for several years, and the platform reports that over three billion pieces of content already carry AI labels thanks to built-in detection and labelling tools. Meta has rolled out its own “AI Info” label for AI-powered posts on Instagram and Facebook, while Google has signed the EU’s voluntary AI transparency code of conduct and is collaborating with major industry players including Nvidia, OpenAI and Apple to develop standardized digital tagging tools.
Even as companies prepare for compliance, some industry leaders warn that overlapping regulatory requirements could create unintended confusion for end users. Karen Massin, a policy lead at Google, warned that excessive regulatory complexity could prove counterproductive to the rules’ core goal, noting that if online platforms become flooded with overlapping AI labels and legal disclosures, it will become harder rather than easier for users to access the clear context they need to evaluate content.