New transparency requirements under Article 50 of the EU AI Act came into force on 2 August. They affect chatbots, AI-generated content, deepfakes, and some biometric AI systems.  

Most of the requirements now apply, although generative AI systems placed on the market before 2 August 2026 have until 2 December 2026 to meet the machine-readable marking requirement. 

What your organisation needs to do depends partly on whether you provide an AI system or use one developed by someone else. Here’s five things AI teams should check now: 

When People Interact With AI, They Need to Know – Article 50(1) 

Providers of AI systems that interact directly with people need to make sure users know they are dealing with AI, unless this is already obvious from the context.  

For chatbots and AI agents, that information should be provided from the beginning of the interaction. 

Check: Do you provide an AI system that interacts directly with people? If so, are users told from their first interaction that they are dealing with AI? 

AI-Generated Content May Need to Be Detectable – Article 50(2)  

Providers of generative AI systems generally need to ensure synthetic text, audio, image, and video outputs are marked in a machine-readable format so they can be detected as AI-generated or manipulated. 

The Act points to techniques such as watermarks, metadata, and digital fingerprints as possible ways of making content detectable. 

This is where the distinction between providing and using an AI system becomes important. A company using a third-party tool such as ChatGPT is not automatically responsible under this provision for adding its own watermark or metadata to every output. 

Check: Do you provide a generative AI system or use somebody else's? If you are the provider, can its outputs be identified as AI-generated? 

Using Emotion Recognition? People Need to Know – Article 50(3) 

Organisations deploying emotion recognition or biometric categorisation systems need to inform the people exposed to them. 

That includes systems designed to infer emotions from biometric data, as well as systems that use biometric data to assign people to particular categories. 

Check: Are you using either of these technologies, and do the people exposed to them know? 

Deepfakes Must Be Disclosed – Article 50(4) 

Organisations using AI to generate or manipulate images, audio or video that constitute a deepfake need to disclose that the content has been artificially generated or manipulated. 

The Act includes special provisions for artistic, creative, satirical, or fictional works, where disclosure can be made in a way that does not hamper the display or enjoyment of the work. 

Check: If you publish AI-generated or manipulated media, could it count as a deepfake? If so, is that clear to the audience? 

AI-Generated Public-Interest Content May Need a Label – Article 50(4) 

The Act also covers AI-generated or manipulated text published to inform people about matters of public interest. 

There is an important exception: disclosure is not required where the content has undergone human review or editorial control and a person or organisation holds editorial responsibility for its publication. 

Check: If you publish AI-generated text on matters of public interest, does it need a disclosure, or does it fall under the editorial-control exception? 

What AI Teams Should Check Now 

Start with the AI systems your organisation already provides or uses. 

For each system, establish whether you are acting as the provider or deployer. Then identify which Article 50 requirements apply and whether your current interfaces, processes and content meet them. 

For many teams, getting that distinction right will be the most important first step. 

The European Commission has also published implementation guidelines and a voluntary Code of Practice on the transparency of AI-generated content to help organisations understand and demonstrate compliance. 

This article provides general information and does not constitute legal advice.