Clarity stands at the core of the European Union’s AI Act, which establishes harmonised transparency duties for AI systems that interact with people, analyse emotions or biometrics, and generate or manipulate synthetic content, including deepfakes. These duties apply broadly to interactive chatbots, virtual assistants, emotion recognition tools, biometric categorisation systems, and generative models that produce text, images, audio, or video for use in public-facing or professional contexts. The obligations are framed as operational requirements for all actors in the AI value chain, including providers, importers, distributors, deployers, and relevant oversight bodies.
The core transparency rule is straightforward: where an AI system is intended to interact directly with natural persons, users must be clearly informed that they are engaging with an AI rather than a human being. This disclosure must occur at or before the first interaction and be presented in a clear and distinguishable manner, avoiding ambiguity about the artificial nature of the system. Additionally, these global standards are crucial for ensuring consistent AI governance across jurisdictions.
Parallel duties apply when deployers expose individuals to emotion recognition or biometric categorisation, requiring prior notice about the operation of the system and its impact on personal data processing.
Synthetic content and deepfakes trigger additional labelling requirements. Providers of generative AI must guarantee that outputs are marked in a machine-readable and detectable way so that downstream users can identify that content as artificially generated or manipulated. Deployers who publish AI-generated text to inform the public on matters of public interest, or who use deepfakes professionally, must disclose the artificial origin of such content in a clear and timely fashion.
Limited exceptions exist for certain law-enforcement uses of biometric categorisation or emotion recognition, subject to strict legal authorisation and safeguards.
Non-compliance with these transparency duties exposes operators to a significant mid-tier of administrative fines. Article 99 of the AI Act sets three main levels of sanctions, with the middle tier allowing penalties of up to €15 million or up to 3% of the preceding financial year’s total worldwide annual turnover, whichever is higher. In addition, Member States are required to provide annual reporting to the Commission on the administrative fines they impose, reinforcing the effectiveness of these penalty rules.
This tier generally applies to violations of obligations governing limited-risk and transparency-only AI systems, distinguishing them from the higher penalties reserved for prohibited practices and the lower tier for minor information failures.
Article 50 obligations, including chatbot disclosure and deepfake labelling, fall squarely within this €15 million or 3% sanction bracket. National market surveillance authorities in each member state are responsible for investigating breaches and imposing fines, guaranteeing consistent enforcement across the single market.
For many businesses relying on customer-facing automation or content generation, this creates substantial financial exposure, particularly for large undertakings with high global turnover.
General-purpose AI model providers face a similar ceiling. For GPAI models, the AI Act empowers EU institutions to levy fines of up to €15 million or 3% of worldwide annual turnover when providers intentionally or negligently infringe applicable requirements, such as transparency, documentation, or cooperation duties.
Failure to provide requested information, documents, or access for evaluation can also trigger this enforcement level, underscoring that opacity around powerful models is treated as a serious regulatory breach rather than a minor compliance lapse.
