EU AI Act model rules hand Brussels new powers from today

EU AI Act rules for AI models became enforceable today, giving Brussels direct authority over general-purpose systems sold in Europe.

The change puts the European Commission at the center of global AI supervision just as large language models are being updated at high speed. The law, passed in 2024, now applies key provisions covering model transparency, copyright disclosures and risk controls.

ChatGPT changed the rulebook

The AI Act began as a framework for AI applications, but the public release of ChatGPT in 2022 shifted the political debate. EU lawmakers widened the law to cover the underlying models that can be adapted for many tasks, rather than only the products built on top of them.

For developers, the new obligations focus on how models are built and how they are explained to business users. Companies must provide information on model capabilities, disclose whether copyrighted material was used in training and give downstream users enough detail to understand operational limits.

The most powerful systems face extra scrutiny. Developers of frontier models must identify broad risks and take steps to reduce potential harm to society, a standard that will test both technical judgment and regulatory patience.

AI Office faces talent squeeze

The Commission created the European AI Office to enforce the model rules, giving it a mandate over one of the fastest-moving parts of the technology economy. Its challenge is not only legal authority, but access to enough technical expertise to assess systems built by some of the world’s richest companies.

Brussels is trying to supplement its own capacity through a panel of scientists and a pool of specialist AI safety firms. That approach reflects a practical constraint: public agencies must compete with private AI labs for scarce talent while the technology itself changes every few months.

The Commission last year endorsed a voluntary code of practice shaped by outside experts, including Yoshua Bengio, to guide compliance. Most leading Western AI labs signed it, while Meta did not; OpenAI’s Tom Duff Gordon said, "We've collaborated closely with the European Commission and the wider ecosystem on implementing the AI Act, including its Codes of Practice, and will continue working together to help Europe realise the benefits of the Intelligence Age."

The political risk is transatlantic. The Trump administration has taken an assertive line against EU digital rules when they affect American technology companies, and enforcement against AI model developers could become another pressure point between Washington and Brussels.

Three paths for model policing

If early enforcement is measured and companies meet documentation demands, the macro effect would likely be a more stable compliance channel for cross-border AI trade. OpenAI and other signatories would gain a clearer operating path in Europe, while the wider industry could treat the EU code as a de facto template for model disclosures.

If Brussels moves quickly against a major lab, the mechanism changes from compliance building to legal confrontation. Macro risk would center on digital trade friction with the United States; affected companies could face fines, product delays or additional reporting demands, while smaller AI developers may struggle to absorb the same compliance burden.

A third path depends on the technology outrunning the rulebook. If frontier models advance faster than regulators can build assessment standards, the global effect would be uncertainty over safety benchmarks, the company-level effect would be less predictable audits, and the sector effect would be rising demand for external AI safety firms.

The open questions are how strict the AI Office will be, how companies document copyrighted training inputs and whether Washington treats enforcement as regulation or commercial retaliation.