New Transparency Obligations Apply Across EU

On Sunday 2 August the first substantive transparency provisions of the EU AI Act entered into force across all member states. The rules cover a wide range of systems, from chatbots and voice assistants through call centres that use emotion recognition to the generation and alteration of images, audio, video and text. The aim is to limit deception and manipulation and to enable people to make informed decisions.

Scope: Chatbots, Emotion Recognition, Deepfakes

Providers and users of AI systems must now clearly inform individuals when they are communicating with a chatbot, virtual assistant or other automated system. So‑called deepfakes, content created or substantially modified by AI that can appear authentic, must also be labelled. For text, the labelling obligation applies when the material is published to inform the public on matters of public interest. The Czech Telecommunications Office (ČTÚ), which will oversee compliance in Czechia, recommends using the distinct labels “AI generated” for fully generated content and “AI modified” for altered content.

The European Commission has issued supplementary guidance stating that AI agents that independently reserve, order or handle correspondence must disclose not only that they are artificial intelligence but also the name of the person on whose behalf they act. The Commission explicitly classifies advertising visuals, product photographs and virtual influencers as deepfakes when they can deceive about the actual appearance or properties of a product. The duty to inform also applies to systems for emotion recognition and biometric categorisation.

Enforcement and Penalties

Fines for breaches of the transparency rules can reach €15 million or 3% of a company’s global annual turnover, whichever is higher. Oversight of model developers will be carried out by the newly created AI Office of the European Commission; the affected firms include not only OpenAI, Anthropic and Google DeepMind but also companies such as Spotify with its AI recommendations and Adobe with AI features in Photoshop. As lawyer Thibau Duquin of Stibbe told Wired: “When you look at it, it’s everywhere, and all of this content will now have to be labelled.”

Implementation Guidance and Transition Period

Organisations should first determine their role, whether they are a provider of a system, a user, or both, and then assess the specific obligations that follow. The ČTÚ therefore recommends that entities adapt their processes in good time. A transition period until December has been granted for the machine‑readable labelling of synthetic content, reflecting the technical complexity of implementation. Critics warn of the risk of “banner blindness”, an excessive number of warnings may lead people to ignore them.

In the Czech Republic the Czech Association for AI (ČAUI) has published a free practical guide to help companies and organisations understand and implement the new rules. As with the GDPR in 2018, authorities are expected to issue further concrete methodology because many terms remain unclear.

Implications for Organisations

Companies must immediately audit their AI systems and processes, establish clear labelling and information duties and prepare for oversight by the ČTÚ. Those who underestimate the requirements risk fines in the tens of millions of Czech crowns as well as reputational damage. The ČAUI guide and forthcoming regulatory methodologies should be the first step towards compliance.