The EU’s AI Act rules on large language models and other foundational AI systems became enforceable on August 2, 2026, establishing the European Commission as the world’s most prominent AI regulator. The law, first passed in 2024, represents the first comprehensive regulation of artificial intelligence technology.
The rules apply to AI models lacking specific purposes but adaptable to various use cases. According to the source, developers must disclose how models were built, reveal any copyright-protected content used for training, and provide information allowing downstream users to understand the model’s capabilities. Companies developing the most powerful “frontier” models face additional requirements to identify and mitigate risks to society.
Last year, the Commission endorsed a voluntary code of practice drafted by leading experts, including Yoshua Bengio. Most major Western AI labs signed the code, with Meta as a notable exception. OpenAI’s Tom Duff Gordon stated the company has “collaborated closely with the European Commission and the wider ecosystem on implementing the AI Act.”
The Commission established the European AI Office to enforce the rules, facing substantial challenges. The EU acknowledges limited resources and competition with the private sector for AI talent. The Commission is tapping external expertise through a panel of scientists and specialized AI safety firms. Officials must navigate fast-moving technological developments without extensive prior experience or scientific consensus on harm prevention.
For European consumers and businesses, the regulation may delay launches of advanced AI models in the EU by weeks as companies ensure compliance. However, the rules theoretically guarantee that AI models available in the EU meet safety standards. The law applies to foreign companies commercializing AI in the EU.
The enforcement approach taken by Brussels is expected to influence global AI regulation, given the EU’s track record with digital rules affecting major tech companies. The Commission faces a choice between focusing on existential risks—such as AI enabling cyberattacks or escaping human control—or addressing the broader spectrum of risks including discrimination and privacy violations. Laura Lazaro Cabrera of the Center for Democracy & Technology cautioned against enforcement becoming “headline-driven,” arguing it should address fundamental rights and societal risks comprehensively.
Key facts
- The EU AI Act rules on large language models became enforceable on August 2, 2026
- Developers must disclose training methods, copyright-protected content, and model capabilities
- Frontier model makers must identify and mitigate societal risks
- The Commission created the European AI Office to enforce the regulations
- Most major Western AI labs signed a voluntary code of practice, except Meta
- The law may delay some advanced AI model launches in the EU by weeks
- The EU’s enforcement approach is expected to influence global AI regulation