The European Union’s AI content labeling rules took effect on 2 August, requiring companies to mark realistic content made or altered by artificial intelligence with visible and machine-readable signals.
The measure sits inside the EU AI Act, the first broad legal framework for the technology. Its aim is to cut misinformation and give people a clear signal when a machine, not a person, produced what they see or read.
The AI content labeling rules reach across formats. Companies must tell users when they interact with an AI chatbot or view synthetic media built to look real. Providers of generative systems must embed markers so images, audio, video, and text can be detected as AI-generated content. Text published to inform the public on matters of public interest also needs a clear label.
The duty splits in two. Firms that build generative systems embed the machine-readable marks. Those that deploy the output must disclose it, above all when the content could pass for real.
How the marking works
For most formats, the mark works on two levels. A watermark sits inside the content, and signed metadata travels with it. Plain text is treated differently and does not carry the watermark. Detection tools can then flag the material as artificially generated or changed. Fines under the AI content labeling rules are now a reality.
Penalties are steep. Breaches can draw fines of up to €15 million or 3% of a company’s total worldwide annual turnover, whichever is higher. For deepfakes, the duty is direct. Anyone using AI to create one must disclose that the content was generated or manipulated.
The AI labeling requirements apply to chatbots, virtual assistants, and any system meant to interact with people. Such systems must be built so users know they face a machine.
Exemptions and grace period
The law carves out clear exceptions. Artistic, satirical, and fictional works stay outside the mandate, as does material made by individuals for personal use. A private group-chat joke is safe. Creative work still carries a lighter disclosure, one shaped so it does not spoil the piece.
One carve-out matters for publishers. AI-written text escapes the labeling duty when a person with real editorial responsibility reviews it and stands behind it. An editor who checks and approves an AI draft can meet that bar.
Developers of existing AI systems get a four-month window to reach full compliance. New systems placed on the EU market face the 2 August date now.
AI transparency rules and public trust
To help firms apply the AI transparency rules, the European Commission published guidelines and a voluntary Code of Practice on the transparency of AI-generated content. Independent experts drew up the code with input from hundreds of stakeholders. Following it is optional. The underlying Article 50 duties are law.
Henna Virkkunen, the Commission’s Executive Vice-President for Tech Sovereignty, Security and Democracy, said the guidelines support the smooth application of the AI Act and help citizens recognise when they deal with AI. She tied the work to building trust and giving innovators firmer ground.
The AI content labeling rules arrive as some technology firms question the wide scope of content that needs a mark. Those same firms back the broader effort against AI-driven misinformation. A separate simplification package could push the machine-marking deadline later in the year, though the core obligations apply now.





