The latest step in the AI Act’s implementation seeks to improve transparency around users’ interaction with the technology and consumption of content produced by it
Transparency rules apply to both providers and deployers of AI systems, but not those using the technology in a personal capacity
On 2 August 2026, key transparency rules under the EU’s AI Act came into force, requiring certain AI systems to inform users when they are interacting with AI and not a human, and when content has been generated or altered by AI. While the act entered into force on 1 August 2024, its provisions have been applied in stages, with different obligations taking effect at different times. Rules applying to high-risk systems in areas such as biometrics, education and critical infrastructure have been delayed by over a year and will now become effective from 2 December 2027, while those regarding high-risk AI embedded into physical products (such as machinery and toys) will apply from 2 August 2028.
The transparency obligations, under Article 50 of the AI Act, apply to providers and deployers of AI, and are intended to reduce deception and manipulation. The new rules require providers to design AI systems in a way that ensures users are explicitly informed when they are directly interacting with an AI system (such as an AI chatbot, agent or avatar), and add machine-readable marks to enable the detection of AI-generated or -manipulated content. Deployers of AI systems must inform individuals when they are exposed to emotion recognition and biometric categorisation tools, deepfakes and AI-generated text on matters of public interest without human review or editorial control. There is an exception for personal, non-professional usage – for example, individuals using an AI system for personal use to generate deepfakes posted on social media would not need to apply labels. However, these deepfakes should be marked in a machine-readable manner by the provider of the AI system to ensure transparency.
The EC has provided optional icons for labelling AI-generated content
The EC has developed a set of freely available icons that deployers of generative AI systems may use to label their AI-generated content. The icons are intended to help those exposed to AI-generated content to recognise that it has been artificially generated or manipulated, to reduce the risk of misinformation. Deployers are required to disclose AI-generated image, audio or video content that resembles existing people, objects, places, entities or events and would falsely appear to be authentic, and AI-generated text published to inform on matters of public interest that has not undergone human review or editorial control. While the use of these specific icons is optional (except in the case of AI-generated or -manipulated content), labelling AI-generated content as such is not.
Guidelines and a code of practice aim to help organisations comply with the new requirements
National market service authorities (i.e. regulators), the European AI Office and the European Data Protection Supervisor (EDPS) are responsible for enforcing the transparency rules. They may issue fines for non-compliance of up to €15m (£12.8m), or 3% of global annual turnover, for companies and up to €750k (£642k) for EU institutions, bodies and agencies, while taking proportionality into account for small- and medium-sized enterprises (SMEs). To help organisations’ compliance efforts, the EC has published guidelines clarifying the new transparency obligations, providing examples of when providers or deployers would have to label content as AI-generated and when they may be exempt. The guidelines explain how compliance could be demonstrated, including through adherence to the voluntary Code of Practice on Transparency of AI-generated Content, which was drawn up by independent experts in a multi-stakeholder process facilitated by the AI Office. At the end of July 2026, around 190 organisations had signed up to the framework, including several prominent AI and tech companies such as Anthropic, Google, Meta, Microsoft and OpenAI.
