EU AI transparency guidelines take effect on August 2: Chatbots need to "self-report their identity" and deepfake content must have machine-readable labels

IT House News on July 26, foreign media Notebookcheck reported today that Article 50 of the EU’s Artificial Intelligence Act will officially take effect on August 2. Under the bill, chatbots must clearly state that they are not real people but AI systems when serving humans; deepfake content must also be clearly stated when posted. Specifically, Article 50 of the Artificial Intelligence Act requires manufacturers to add obvious markings when designing chatbots and voice assistants so that users can know that they are communicating with machines. AI-generated images, audio, video, and text must be accompanied by machine-readable tags declaring themselves to be AI content. At the same time, the bill requires users and institutions to provide clear explanations when publishing deep fake (IT House Note: Deepfake) content, and AI-generated text involving public interests must also be labeled. However, AI spell checking and grammar correction tools are exempt and do not need to be marked, and content with obvious fantasy properties such as "free flying humans" does not fall into the category of deep fakes. At the same time, “machine-readable” markings under the bill do not mean visible to the human eye. Although the EU provides three "AI-generated" labels for public use, these labels are essentially voluntary. The bill only requires that AI-generated content must be marked with watermarks, metadata and other information for detection tools to read. In addition, the bill also provides for a transition period. Generative AI systems that have been launched before August 2 will be given 4 months to rectify, with a deadline of December 2, 2026. Companies that breach the Act can be fined up to €15 million, or 3% of overall turnover.