The European Union will gain enforcement power from Sunday over landmark artificial intelligence rules, allowing regulators to act against companies that break the AI Act and impose fines. The EU adopted the AI Act in 2024, introducing its rules in stages as part of a broader digital rulebook covering social media, online retailers and search engines.

The AI Act uses a risk-based approach: the higher the risk to Europeans' rights or health, the greater the obligations on firms to protect people from harm. The law bans AI used for predictive policing, emotion recognition systems in workplaces or schools, and AI that manipulates human behaviour. From Sunday, AI-generated content such as deepfakes must be labelled, and systems such as chatbots must make clear to users that they are AI. Existing AI systems have until December 2 to adapt, with exemptions for "artistic, creative, satirical, fictional" work.

European regulators in the AI Office, a body attached to the EU executive and staffed by dozens of tech experts, lawyers and economists, will be able to enforce rules on the most advanced general-purpose AI models (systems usable across many tasks), including chatbots. The AI Office can test models for compliance, require companies to give the EU access to their models, and in some cases demand access before models are placed on the market, according to an EU official. The EU can also restrict a model's deployment in the bloc if concerns are not addressed by the provider.

The article says the new powers follow EU difficulty this year in getting access to American AI developer Anthropic's Mythos model. National authorities will enforce rules for smaller AI systems. Companies that allow banned practices face fines of up to seven percent of annual worldwide turnover or 35 million euros ($40 million), whichever is higher. Other violations can bring fines of up to three percent of global annual revenue or 15 million euros, including cases where firms hinder or fail to cooperate with EU probes.

More AI Act rules will become enforceable later this year and beyond. From December, the EU will ban AI systems that generate sexualised deepfakes, following global outrage over non-consensual nudes produced by Elon Musk's chatbot Grok. Some high-risk AI rules for models deemed potentially dangerous to safety, health or citizens' fundamental rights are delayed to give companies more time: stand-alone high-risk AI systems face rules from December 2027, while AI tools embedded in other products face rules in August 2028.

The European Union will gain enforcement power from Sunday over landmark artificial intelligence rules, allowing regulators to act against companies that break the AI Act and impose fines. The EU adopted the AI Act in 2024, introducing its rules in stages as part of a broader digital rulebook covering social media, online retailers and search engines.

The AI Act uses a risk-based approach: the higher the risk to Europeans' rights or health, the greater the obligations on firms to protect people from harm. The law bans AI used for predictive policing, emotion recognition systems in workplaces or schools, and AI that manipulates human behaviour. From Sunday, AI-generated content such as deepfakes must be labelled, and systems such as chatbots must make clear to users that they are AI. Existing AI systems have until December 2 to adapt, with exemptions for "artistic, creative, satirical, fictional" work.

European regulators in the AI Office, a body attached to the EU executive and staffed by dozens of tech experts, lawyers and economists, will be able to enforce rules on the most advanced general-purpose AI models (systems usable across many tasks), including chatbots. The AI Office can test models for compliance, require companies to give the EU access to their models, and in some cases demand access before models are placed on the market, according to an EU official. The EU can also restrict a model's deployment in the bloc if concerns are not addressed by the provider.

The article says the new powers follow EU difficulty this year in getting access to American AI developer Anthropic's Mythos model. National authorities will enforce rules for smaller AI systems. Companies that allow banned practices face fines of up to seven percent of annual worldwide turnover or 35 million euros ($40 million), whichever is higher. Other violations can bring fines of up to three percent of global annual revenue or 15 million euros, including cases where firms hinder or fail to cooperate with EU probes.

More AI Act rules will become enforceable later this year and beyond. From December, the EU will ban AI systems that generate sexualised deepfakes, following global outrage over non-consensual nudes produced by Elon Musk's chatbot Grok. Some high-risk AI rules for models deemed potentially dangerous to safety, health or citizens' fundamental rights are delayed to give companies more time: stand-alone high-risk AI systems face rules from December 2027, while AI tools embedded in other products face rules in August 2028.