In the European Union, starting on August 2, new requirements for labeling content created by artificial intelligence (AI) and marking deepfakes will take effect. These measures are part of the transparency provisions of the comprehensive AI Act. The new rules will provide users in the EU with more information about when they are interacting with artificial intelligence systems.
Under the law, developers and providers of artificial intelligence systems must ensure that users are informed when they are directly interacting with AI. They are also required to include machine-readable labels that identify content created or modified using AI.
Chatbots
Any website, customer support service, or government agency using a chatbot must inform users that they are communicating with an artificial intelligence system rather than a human. An exception is made in cases where this is obvious, such as when users see a message like “ask our AI chatbot for additional information” or similar.
This message must be clear, understandable, and displayed at the appropriate time. The chatbot provider is responsible for ensuring that users receive this information. The new requirements will apply to all chatbots operating in the EU from August 2. Existing services will have a transitional period until December 2026, although many already comply with these requirements.
Content created or modified by AI
The new rules also apply to images, videos, audio, and text created or edited using artificial intelligence. Such content must include a technical label that identifies its artificial origin.
This label can take various forms, including data embedded in the file, information about the origin, or a technical watermark accessible to verification tools. It is essential that this information remains intact even after the file is cropped or otherwise modified. Verification tools must be able to access this data, though it does not need to be visible or understandable to the average user.
As a result, most photographs, videos, or articles created using AI are not required to display a visible label stating “generated by AI”. The same principle applies to original content that has been altered using artificial intelligence.
Deepfakes
Deepfakes, except in certain cases of personal, artistic, satirical, or fictional use, must be clearly labeled as created or modified using AI. If a video features a politician, businessman, or other well-known individual supposedly addressing viewers or making statements, users must be able to recognize that this material was generated or altered by artificial intelligence.
According to the AI Act, an audio deepfake, similar to the fake message from Joe Biden urging Americans not to vote in the 2024 US presidential election, must begin with a warning phrased in “simple and understandable natural language” that the recording was created using AI. Additionally, emotion recognition systems and biometric categorization, including facial recognition, must inform individuals about the use of such technologies starting on August 2.
Next steps
The next phase of implementing the AI Act, which covers tools for recruitment, credit scoring, education, law enforcement, border control, and critical infrastructure, will take effect on December 2, 2027. These sectors will face significantly stricter requirements.
Systems used as safety components in certain products, including medical devices, toys, or elevators, will be subject to these rules starting on August 2, 2028. Violations of the requirements may result in fines of up to 35 million euros or 7% of the company’s global annual revenue.
Source: The Local Europe
