This page covers transparency, general-purpose AI models and fines from our EU AI Act guide, which also explains the four risk tiers, the implementation timeline and what to do now. The prohibited practices and high-risk rules are in prohibited AI practices and high-risk AI systems.
Article 50 transparency: duties depend on your role
Article 50 has applied since 2 August 2026. It does not require a visible AI label on every piece of AI-generated text, image or internal document. It sets separate duties for providers and deployers, each with its own exceptions. Generative systems already on the market before 2 August 2026 have until 2 December 2026 to add machine-readable marking. Our article on Article 50 labelling and chatbot disclosure covers the detail.
| Who | Main duty |
|---|---|
| Provider of a system that interacts directly with people | Make sure people are told they are interacting with AI, unless that is obvious from the context or another exception applies |
| Provider of a system that generates synthetic audio, images, video or text | Mark the output in a machine-readable format so it can be detected as AI-generated, as far as technically feasible |
| Deployer of an emotion recognition or biometric categorisation system | Inform the people exposed to it |
| Deployer publishing deepfake image, audio or video | Disclose that the content was artificially generated or manipulated, with lighter rules for evidently artistic or satirical work |
| Deployer publishing AI-generated text to inform the public on matters of public interest | Disclose it, unless the text has undergone human review and a person or organisation holds editorial responsibility |
General-purpose AI models: what deployers need to know
General-purpose AI (GPAI) models are the large models behind products such as ChatGPT, Microsoft 365 Copilot and Google Gemini. Those products are AI systems built on GPAI models, not the models themselves. GPAI obligations on model providers have applied since 2 August 2025: technical documentation, information for downstream providers, a copyright policy and a public summary of training content.
The model provider's compliance covers the model provider's duties. It does not pass to you, and it does not show that your use of the system is compliant. If you use these tools, you still assess your own obligations for each use case: data protection, security, transparency, informing workers and, in some cases, high-risk duties. The provider's contractual and technical documentation helps with that assessment.
You can also become a provider yourself. An organisation is the provider of an AI system when it develops that system, or has it developed, and places it on the market or puts it into service under its own name, or when it makes one of the changes covered by Article 25. Fine-tuning or integrating a GPAI model does not on its own make every high-risk provider obligation apply. A company fine-tuning a model for internal experiments is in a very different position from one that builds a CV screening tool on a GPAI model and puts it into use. Assess the purpose, your role and the classification of the resulting system. Using a third-party GPAI model, through an API or inside a product, does not make you the provider of that underlying model; its developer keeps the GPAI model obligations. You may still become the provider of the AI system you build on it. Substantially modifying the model itself, for example through extensive fine-tuning, is a separate question.
GPAI models with systemic risk
Article 51 presumes that a GPAI model has high-impact capabilities when the compute used to train it exceeds 10²⁵ floating point operations (FLOPs). The European Commission can also designate a model as having systemic risk based on other criteria. Providers of these models face additional duties, including model evaluation and adversarial testing, serious incident reporting to the AI Office, and cybersecurity protection. As a deployer, there is no general public register of GPAI providers to check. Instead, review the provider's documentation for downstream users, its published training content summary and its other disclosures.
The fine structure
National market surveillance authorities enforce the Act in each member state. The Commission, through the AI Office, supervises GPAI providers.
- Prohibited practices: up to €35 million or 7% of total worldwide annual turnover, whichever is higher.
- Most other obligations, including high-risk and transparency duties: up to €15 million or 3% of total worldwide annual turnover, whichever is higher.
- Incorrect, incomplete or misleading information to authorities: up to €7.5 million or 1% of total worldwide annual turnover, whichever is higher.
For SMEs, including start-ups, each fine is capped at whichever of the two figures is lower. When setting a fine, authorities weigh the factors listed in the Act, including the nature, gravity and duration of the infringement. Prepare evidence of compliance in line with the obligations and dates that apply to you, rather than assuming early infringements will be tolerated.