From 2 August the EU is supervising artificial intelligence models

Enforcement of the European artificial intelligence law entered a new phase on 2 August. The European Commission, through its AI Office and in collaboration with the competent national authorities, has begun the surveillance and control activities provided for by the AI Act.

From that date transparency obligations apply to chatbots and interactive systems. They must clearly inform those using them that they are dealing with a machine.

A second obligation concerns content. Images, videos and audio files generated or modified with artificial intelligence, including deepfakes, must be labelled and carry machine-readable markers so that they are easier to identify.

General-purpose models and prohibited practices

The AI Office can now enforce the provisions against suppliers of general-purpose models, including advanced ones that present systemic risks.

The obligations set out for these suppliers aim to contain wide-ranging risks. The text cites chemical, biological, radiological and nuclear incidents, loss of control over systems, cybercrime, harmful manipulation and threats to fundamental rights.

All suppliers of general-purpose models must document information for the competent authorities and for developers. They must also adopt copyright compliance policies and publish a detailed summary of the content used for training.

Supervision also covers the practices prohibited by the regulation. These include facial recognition, social scoring tools, emotion recognition and subliminal or manipulative techniques.

The AI Office has set up digital tools for reporting. Citizens, businesses and whistleblowers can confidentially report suspected breaches of the law.

The Commission's stated aim is to reduce the risk of manipulation and to allow citizens to make informed choices. For businesses, the regulation is meant to offer a clear framework for compliance.

The regulation applies in the European Union. Its effectiveness will now depend on whether the AI Office and national authorities can keep pace with a field in which models change faster than administrative procedures.