Add the Gazzetta del Sud as a source
More transparency for those who place on the market or use chatbots, image generators, synthetic audio and video, emotion recognition systems and tools capable of producing deepfakes. These are the main obligations that will become applicable on 2 August 2026, the date on which the transparency obligations provided for by the AI Act will apply in the European Union, the first organic regulation in the world on Artificial Intelligence, in force from 1 August 2024. Violations are punishable by up to 15 million euros or, for companies, up to 3% of the total annual worldwide turnover of the previous financial year, whichever is higher.
The Commission’s guidelines
«The Commission has adopted these guidelines to provide practical guidance to the competent authorities, as well as to suppliers and users of AI systems – it writes in the document on the guidelines published on 20 July – The objective is to ensure compliance with the transparency obligations set out in Article 50 of the AI Act in a coherent, effective, proportionate and uniform way».
The first change is that people will have to be informed when interacting with an artificial intelligence system, unless this is obvious. A useful clarity for users and above all for the most fragile categories such as minors. The synthetic contents must have a machine-readable technical marking paid by the suppliers (provider) and information for users (deployer). An exception is made for texts of public interest subjected to human review or editorial control, for which a natural or legal person assumes editorial responsibility.
Obligations for suppliers and users
In particular, users of artificial intelligence systems (deployers) will have to inform users if “exposed to deepfakes, to content generated by AI on issues of public interest without human review or editorial control and to emotion recognition or biometric categorization systems”. Suppliers will have to ensure, in the case of systems that generate synthetic audio, image, video or textual content, that the technical solutions adopted are “effective, interoperable, solid and reliable as indicated in the relevant technical standards”.
Furthermore, in the case of emotion recognition systems or a biometric categorization system, users are required to inform exposed natural persons, unless the systems are authorized by law for criminal investigations. Furthermore, the user of an AI system that generates or manipulates images or audio or video content that constitutes a deepfake will have to disclose that the content “has been artificially generated or manipulated”, with methods attenuated for artistic, creative or satirical works.
What is left out
They do not fall within the scope of the art. 50: personal non-professional activity; research and development, a sector which excludes systems developed and put into service exclusively for scientific research purposes, as well as research, testing and development activities prior to being placed on the market; the systems authorized by law to ascertain, prevent or investigate crimes, within the limits established by the regulation.
Belisario: «We’re starting to get serious»
«Two years after the entry into force of the AI Act we are starting to get serious about it – Ernesto Belisario, a lawyer specializing in the law of new technologies, explains to ANSA – Transparency on the contents generated by artificial intelligence is necessary but must not be reduced to a bureaucratic requirement, and therefore not very useful, as happened with the cookie information or on the processing of personal data. However, quicker times would be needed: guidelines published less than two weeks before the deadline leave companies with very narrow margins to adapt to technologies that evolve so quickly”.