EU AI transparency rules for chatbots and deepfakes start Aug. 2

The European Commission has issued guidance for AI Act transparency obligations that take effect on Aug. 2, requiring many AI providers and deployers in the EU to tell users when they are dealing with software rather than a person. The rules also require machine-readable labeling for AI-generated or AI-altered content, a compliance change that will affect chatbot vendors, AI agent products, content platforms and companies using synthetic media in public-facing services.

The Commission said the guidance is meant to help companies apply the law before the deadline. Henna Virkkunen, the Commission’s executive vice-president for tech sovereignty, security and democracy, said the measures are intended to make chatbots, AI agents and AI-created content “more transparent and trustworthy,” and to help citizens recognize when they are interacting with AI.

For technology companies, the immediate operational issue is disclosure. Organizations deploying covered systems in the EU will have to identify interactions with AI systems such as chatbots. They will also have to disclose deepfakes to people exposed to them and mark synthetic or manipulated audio, images, video and text in a machine-readable way.

The requirements also cover AI-generated or AI-manipulated text published to inform the public on matters of public interest. There is an exception where that material has been subject to human review or editorial control. The Commission’s guidance also includes emotion recognition systems and biometric categorization systems within the transparency regime.

What is covered

• Interactive AI systems, including chatbots and AI agents.
• Synthetic audio, images, video and text.
• Deepfakes shown to or otherwise encountered by people in the EU.
• AI-generated or AI-altered text used to inform the public on public-interest matters, unless it has human review or editorial control.
• Emotion recognition and biometric categorization systems.

The rules do not apply to systems carrying out ordinary editing functions, such as spelling or grammar correction, where the tool does not materially change the user’s input. That carveout matters for software vendors adding AI features to productivity tools, although the Commission’s guidance, not vendor marketing, will determine the edge cases.

Aug. 2 is also a broader enforcement milestone. Most of the AI Act will apply from that date, and the Commission will gain authority to enforce the rules for general-purpose AI models. The financial penalties and enforcement mechanics were not detailed in the guidance summary, and companies still have to map which parts of their products fall under provider obligations, deployer obligations or both.

The AI Act entered into force in August 2024 after the Commission first proposed it in 2021, before ChatGPT made generative AI a mainstream policy issue. The legislation created obligations for model developers, companies integrating AI into software products and organizations deploying AI systems.

The rollout has been staggered. Earlier provisions banned certain uses, including biometric categorization systems that purport to classify people by traits such as political views, religion, sexual orientation or race. Rules for general-purpose AI were added late in the legislative process to bring models such as OpenAI’s ChatGPT within scope, and those rules came into force in 2025.

Parts of the law have already slipped. In May, negotiators for the European Parliament and Council agreed to delay some AI Act requirements after months of complaints from industry. Rules for standalone high-risk AI systems are now scheduled for Dec. 2, 2027, while rules for high-risk systems built into regulated products have been moved to Aug. 2, 2028.

This story draws on original reporting from The Register.