EU’s AI Act kicks in, deepfake labels must under new rules

The European Union on Sunday began enforcing key provisions of its landmark Artificial Intelligence (AI) Act, bringing providers of advanced AI models under regulatory oversight and making it mandatory for chatbots and AI-generated content to clearly disclose their artificial origin.

The move marks the first major enforcement phase of the world’s first comprehensive AI law as the European Commission seeks to curb misuse of rapidly evolving artificial intelligence technologies while boosting public trust in their deployment. From August 2, the European Commission’s AI Office, alongside national authorities, assumed enforcement powers over general-purpose AI (GPAI) models that power a wide range of applications, including chatbots and AI agents.

Under the new rules, chatbots and other AI systems interacting with users must inform them they are communicating with AI rather than a human. AI-generated or AI-altered images, videos and audio, including deepfakes, must also be labelled and carry machine-readable markers to make them easier to detect.

The commission said the measures were intended to reduce deception and manipulation, help users make informed choices and provide businesses with a clear compliance framework. It also published a list of more than 180 organisations that have signed a voluntary Code of Practice on transparency of AI-generated content.

Providers of advanced AI models posing systemic risks will face additional obligations to mitigate threats ranging from cyberattacks and harmful manipulation to chemical, biological, radiological and nuclear risks.

All GPAI providers must also maintain technical documentation, adopt copyright policies and publish summaries of the material used to train their models.

The commission also launched complaint and whistleblower mechanisms to report suspected violations and appointed Oxford University professor Alessandro Abate as Lead Scientific Adviser.