BRUSSELS, (Belgium): Starting this Sunday, the European Union will officially enforce its new artificial intelligence rules, giving it the authority to impose penalties on companies that fail to comply.
The EU is known for having one of the most detailed sets of digital regulations in the world, covering areas such as social media, online sales, and search engines. More regulations are expected to follow within the year.
In 2024, the EU rolled out an extensive AI law known as the “AI Act,” which has been introduced gradually.
Here’s what to expect on Sunday:
The EU’s approach focuses on assessing risks associated with AI technology. The greater the potential risk to individuals’ rights or health, the more responsibility businesses have to safeguard people from possible harm.
Certain uses of AI will also be prohibited, including predictive policing, emotion recognition in schools and workplaces, and AI systems designed to manipulate user behavior.
From Sunday onwards, AI-generated content, like deepfakes, will need clear labeling. Companies will also have to ensure that their AI systems, such as chatbots, make it obvious to users they’re interacting with AI.
Existing AI systems will have until December 2 to comply with these new rules, although creative works, including art and satire, may be exempt.
On Sunday, the newly established AI Office in Europe will start enforcing regulations on advanced AI models. This regulatory body will consist of a variety of tech experts, lawyers, and economists.
The AI Office will have the capacity to test advanced general-purpose AI models, like chatbots, to confirm that they meet legal requirements. Companies will need to provide the EU with access to their models, and the EU can request this access even before the products hit the market.
If the EU has concerns that a model could pose risks, it reserves the right to restrict its use within Europe.
These regulations come at a crucial time, as the EU faced challenges obtaining access to Anthropic’s AI model earlier this year. With these new rules in place, it aims to avoid such issues in the future.
National authorities will also have the ability to enforce these rules for smaller AI systems. If companies are found violating the regulations, the EU can compel them to fix the issue and may impose fines for non-compliance.
The largest fines can reach up to seven percent of a business’s annual global revenue or 35 million euros (about $40 million), whichever is higher. For other offenses, fines could be up to three percent of global revenue or 15 million euros, with additional penalties for obstructing EU investigations.
More regulations under the AI Act will roll out later this year and beyond. Starting in December, there will be a prohibition on AI systems generating sexualized deepfakes, a move prompted by public backlash over non-consensual content produced by a recent AI model.
The EU is also postponing the launch of certain high-risk AI rules that involve models considered potentially harmful to safety, health, or citizens’ rights, providing businesses with more time to adapt. These regulations are set to take effect for high-risk AI systems by December 2027 and for AI tools integrated into other products by August 2028.
