Source: Taipei Times

From today the European Union’s executive gains enforcement powers under its landmark AI rulebook, backed by a newly empowered AI Office attached to the European Commission that will include tech experts, lawyers and economists. That office can test the most advanced general-purpose AI models, demand access to models (including, in some cases, before market placement), and restrict deployment when provider responses do not resolve regulators’ concerns. National authorities will handle enforcement for smaller AI systems while the AI Office focuses on cutting-edge models.

The regime uses a risk-based approach: it bans AI for predictive policing, emotion-recognition in workplaces or schools and systems that manipulate human behavior, and requires labeling of AI-generated content such as deepfakes and clear disclosure when users interact with chatbots. Existing AI systems must adapt by Dec. 2, and a ban on AI-generated sexualized deepfakes will take effect from December; some high-risk obligations have been delayed (standalone high-risk systems from December next year, embedded AI tools from August 2028) to give companies more time. The rules include stiff fines — up to 7% of annual worldwide turnover or €35 million for banned practices, and up to 3% or €15 million for other contraventions — and penalties for hindering or failing to cooperate with EU probes. The move follows difficulties this year in obtaining access to Anthropic’s Mythos model and signals further AI-related rules to come.