The wait is over.
The EU’s transparency obligations under Article 50 of the EU AI Act are now in force.
To support this implementation, the European Commission published its Guidelines on Transparency Obligations for Providers and Deployers of Certain AI Systems on 20 July 2026, which are non-binding guidelines to assist providers, deployers and competent authorities on how to comply with Article 50 and explain the four distinct transparency obligations under the AI Act.
Adherence to an appropriate code of practice, such as the Code of Practice on Transparency of AI-Generated Content, can also demonstrate compliance.
Companies should take action to ensure that their AI systems, internal policies and customer-facing disclosures meet the new requirements. Providers of legacy AI systems that generate synthetic audio, image, video or text content and were placed on the market before 2 August 2026 have until 2 December 2026 to comply with the marking and detection obligations under Article 50(2). All other transparency obligations under Article 50, however, went into force on 2 August 2026.
Non-compliance with the transparency obligations may result in fines of up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher.
The key points to be aware of are summarised below, together with practical examples drawn from our experience advising clients on AI compliance.
Disclosure of direct AI interaction (Article 50(1))
Providers of AI systems designed to interact directly with individuals must ensure that users are clearly informed they are interacting with an AI system, unless this would be obvious to a reasonably well-informed person. For example, an AI-powered chatbot handling customer queries must display a notice, such as “You are interacting with an AI system”, at the outset of the conversation, supplemented by a visible AI icon.
Similarly, an AI-operated telephone hotline must provide a verbal disclosure at the start of the call with periodic reminders during longer interactions. Exceptions include simple automated responses (e.g. traditional out-of-office replies), AI tools that merely assist a human representative who delivers the final response, and internal employee-facing AI assistants used by properly trained AI-literate staff.
Machine-readable markings of AI-generated content (Article 50(2))
Providers of AI systems that generate synthetic audio, image, video or text must mark that content in a machine-readable format, using technical solutions that are effective, interoperable, robust and reliable, so that it is detectable as artificially generated or manipulated. For instance, AI-generated marketing visuals, promotional videos or an AI-composed tune for an advertisement must all carry such machine-readable markers. A key exception covers standard editorial assistance that does not substantially alter the original content’s meaning, such as AI-assisted translation, format conversion, minor cropping, colour correction or background removal.
Notice for emotion recognition and biometric categorisation (Article 50(3))
Deployers of emotion recognition or biometric categorisation systems must inform the affected individuals of the system’s operation before or at the point of first exposure. For example, deploying facial-analysis technology in a retail environment for age estimation requires a clear notice at the entrance explaining the system’s function. Equally, an emotion- or tone-recognition feature on a customer service voice line used to analyse callers’ mood and route calls must inform customers.
Labelling deep fakes and AI-generated public-interest text (Article 50(4))
Deployers who use AI to generate or modify deep fake image, audio or video content or text published to inform the public on matters of public interest must clearly and perceptibly disclose that the content is AI-generated or manipulated, no later than first exposure. For example, an AI-generated advertisement featuring a realistic synthetic depiction of an influencer or an AI-modified product image that could create a false impression of the product’s actual appearance must be labelled as AI-generated. Similarly, an AI-generated summary of a corporate report published on an investor relations page must carry such a disclosure.
A notable exception applies where the text has undergone genuine human editorial review and control with a natural or legal person bearing editorial responsibility for the publication. Purely formal checks do not satisfy this criteria.
For more information on the EU AI Act and its transparency obligations, contact your CMS client partner or the CMS experts who contributed to this article.
