New European Union transparency rules will require companies to label content generated or modified by AI, including responses, images, audio and video from chatbots, starting Sunday, August 2.
Companies operating in the EU must use both visible labels and technical markers, such as watermarks or embedded metadata, to identify synthetic material. Existing AI systems must comply by December 2, after which enforcement will increase and significant fines may be imposed.
This requirement is part of the broader EU AI law, which is being implemented in phases. The initial phase focuses on dissemination.
What EU AI labeling rules require
The rules require clear identification of content generated or significantly altered by AI. This includes both on-screen labels and technical markers, such as watermarks or embedded metadata, to identify synthetic material.
The requirements go beyond the labels on the screen. Companies should implement systems that identify AI-generated content even after it is shared or reposted, using watermarking and tagging tools that persist across platforms. Regulators are concerned about the rapid spread of compelling AI-generated media and the increasing difficulty in distinguishing real from synthetic content.
These requirements primarily apply to content created in professional contexts. Public interest material produced by AI without human editorial oversight must be labeled.
The EU has established several exceptions. People who use AI for personal purposes are not covered, and exemptions apply to “artistic, creative, satirical and fictional” works. These exceptions limit the rules to professional and public information content.
How the main platforms are responding to the new rules
Several major technology companies have already taken similar measures. TikTok has required creators to tag AI-generated content for several years and reports that more than three billion pieces of content have been tagged using its detection tools. Meta has introduced an “AI Insights” tag on Facebook and Instagram to mark posts created with generative AI.
Google has signed the EU voluntary code of conduct on AI transparency and is collaborating with companies such as Nvidia, OpenAI and Apple to develop digital labeling standards to trace the origins of content.
Some companies are concerned about the launch, arguing that the rules add complexity to already crowded platforms. Google’s Karen Massin said more regulatory complexity could backfire and confuse users, and noted that constant labels and disclosures could lose their impact if users encounter too many overlapping indicators.
Others see this pushback as typical of new compliance requirements. “We’ve heard that it’s going to be very, very difficult to implement. But I think we hear this often regarding compliance requirements. And yet the world turns and we realize these things,” Ashley Casovan of the International Association of Privacy Professionals told AFP.
What companies should do before December 2
Companies operating in the EU that produce or host AI-generated content must assess their compliance by the December 2 deadline for existing systems:
- Identify what content falls within the professional and public information realm, as opposed to artistic and personal use exemptions.
- Implement visible labels on images, audio, video, and chatbot output generated or significantly altered by AI.
- Add persistent technical markers, such as watermarks or embedded metadata, that remain intact when content is shared or republished across platforms.
- Review whether human editorial oversight is applied to AI-produced public interest content, as this oversight affects labeling obligations.
- Ensure existing AI systems are compliant before December 2, when enforcement actions and fines begin.
The labeling requirements will go into effect on August 2, 2026, with a compliance deadline of December 2 for existing AI systems. Enforcement of the law will increase after that date, with significant fines for non-compliance.
The EU has not specified exact penalties or how it will verify that persistent watermarking and labeling systems meet required standards across platforms.
