The EU AI Act, which requires all member states to adopt a unified AI risk framework and sets regulations for AI developers and deployers, will be gradually implemented over the next two years.
On August 1st, the world's first artificial intelligence (AI) law, the AI Act established by the European Union, officially came into effect. It introduces a unified AI risk framework for all EU countries and outlines clear responsibilities and obligations for AI developers and deployers. The implementation and enforcement of this law will be managed by the AI Office under the European Commission (EC).
The AI Act defines four levels of AI risk and corresponding codes of conduct. Most AI systems are categorized as minimal risk, such as spam filters or AI games, which are not subject to obligations but can voluntarily adopt additional codes of conduct. Next are specific transparency risks, requiring systems like chatbots to clearly inform users that they are interacting with a robot and marking certain AI-generated content.
AI systems classified as high-risk must meet more stringent requirements. Examples include AI-based medical software or AI systems used for recruitment, which must have risk mitigation systems, use high-quality datasets, provide clear user information, and be subject to human oversight.
AI systems that appeared in shows like "Black Mirror," where the entire human society is scored by AI, are deemed unacceptable risks. Social scoring AI systems, whether government or corporate-led, are considered threats to fundamental human rights and are outright banned.
The European Commission stated that the EU aspires to be a global leader in safe AI. By establishing a robust regulatory framework based on human rights and fundamental values, the EU aims to develop an AI ecosystem that benefits everyone.
To assist companies in complying with the AI Act, the EC is working on drafting a code of conduct for General Purpose AI (GPAI) model providers, with the official code expected to be released by April next year.
Although the AI Act is now in effect, the implementation timelines for different rules vary. For instance, the ban on unacceptable risk AI systems will take effect in six months. EU member states must designate supervisory authorities by August 2nd next year to oversee the deployment of AI system rules. Rules for General Purpose AI models will be implemented in 12 months, and the majority of AI Act rules will not come into force until August 2nd, 2026.