From 2 August 2026 a new phase of the AI Act, the European regulation on artificial intelligence, comes into force. From this date the transparency obligations of Article 50 and the rules on general-purpose AI models apply.
The rules concern chatbots, deepfakes and emotion recognition systems. They involve both the providers of the systems and the deployers, meaning those who use them in their business processes.
The same date marks the start of supervision and of the sanctions regime for GPAI models. It is a step that directly affects companies that develop or integrate artificial intelligence.
What GPAI models are and which obligations they have
GPAI models are large foundation models that can be used for multiple applications. From 2 August 2026 their providers must document the technical features of the model.
This information must be made available to the competent authorities and to downstream operators. Providers must also adopt a policy on copyright compliance.
There is also an obligation to publish a summary of the content used to train the model. The aim is to increase traceability and transparency along the value chain.
Transparency obligations for content and systems
Article 50 requires that content generated or substantially manipulated by artificial intelligence be made recognisable. This includes images, audio and video, as well as synthetic voices and deepfakes that portray real people.
The obligation also covers texts published to inform the public on matters of general interest. Systems that interact with people must clearly signal the artificial nature of the counterpart.
The provider is the entity that builds the AI system. It must handle the design of systems that interact with people and the marking of generative outputs.
The deployer, on the other hand, is the entity that uses artificial intelligence in its own processes. This role is responsible for making the information visible and understandable to the end user.
Supervision, fines and what companies must do
From 2 August 2026 the European Commission, through the AI Office, can request information and assess the models. It can also impose corrective measures and sanctions.
Fines for providers of GPAI models can reach up to 15 million euros or 3% of annual worldwide turnover. The sanctions regime makes compliance with the obligations concrete.
For companies the first step is to understand which role they hold, whether provider or deployer. The concrete requirements to be implemented depend on this distinction.
Those who use chatbots or generative tools should check the notice given to the user. It is also useful to review the contract with the software provider to clarify who handles the technical marking and who handles the visible transparency.
The 2 August 2026 deadline therefore calls for a review of the processes that involve artificial intelligence. The adjustment concerns documentation, content labelling and communication with the public.
Source: https://www.ilsole24ore.com/art/ai-act-2-agosto-debuttano-vigilanza-e-doveri-trasparenza-AJtYEMV
Source: agendadigitale.eu
Original article: digitaldomain.it
