U.S. and EU AI Regulation: Parallel Regimes, Not Joint Framework

28 October 2026

There is no evidence that the U.S. and EU have agreed to a joint framework to regulate foundation AI models at a G7 summit. Instead, the two jurisdictions have established parallel but separate regimes, accompanied by non-binding international cooperation initiatives.

EU Launches Pioneering AI Act

In late 2023, EU lawmakers finalized a pioneering legislative framework, the EU AI Act. The act introduces binding rules for general-purpose AI models and creates a new European AI Office for centralized governance.

Under the AI Act, providers of all general-purpose AI models must disclose key information to downstream system providers, aiding their understanding of the underlying models. The act presumes models trained using over 10^25 FLOPs of computing power to pose systemic risks.

The EU AI Act classifies AI systems into risk categories with varying obligations. It directly applies to any provider of general-purpose AI models operating within the EU, unlike U.S. rules that primarily bind federal agencies and certain developers.

U.S. Executive Order Focuses on Dual-Use Risks

In 2023, U.S. President Joe Biden signed Executive Order 14110, focusing on "dual-use foundation models." These are models with high performance at tasks posing security or safety risks. The order requires developers of such models to report plans, cybersecurity protections, and vulnerability test results to the Department of Commerce.

The executive order directs the National Institute of Standards and Technology (NIST) to develop guidelines and standards for testing AI for dangerous capabilities and assessing AI safety, security, and trustworthiness. Like the EU act, it uses compute thresholds to categorize higher-risk models.

However, the U.S. order has a narrower focus on national security and critical infrastructure risks, using existing authority mainly to create reporting requirements. It does not impose direct regulation on private industry beyond those reporting duties.

EU Focuses on Wider Systemic Risks

In contrast, the EU AI Act casts a wider net, addressing risks such as discrimination at scale, major accidents, and negative impacts on human rights. Under the act, GPAI model providers must document training data and capabilities, demonstrate risk-mitigation steps, and undergo external audits.

The act creates two categories for foundation models: higher-risk "systemic" models subject to stricter obligations, and other foundation models. It applies in all EU member states, governing AI development, sale, and use.

International Cooperation: Non-Binding Commitments

The U.S. and EU have also pursued non-binding international cooperation on AI regulation. The EU-U.S. Trade and Technology Council (TTC) adopted a joint roadmap for AI evaluation and measurement tools, including a commitment to common terminology for trustworthy AI.

G7 leaders published the Hiroshima Process International Code of Conduct for Organizations Developing Advanced AI Systems in October 2023. This voluntary guidance encourages actions like risk mitigation, transparency, security controls, and research on societal and security risks.

However, legal analyses note that the G7 code is high-level, lacking concrete enforcement mechanisms and allowing different implementation by jurisdictions. It does not create a single harmonized regulatory framework.

The code could serve as a basis for international AI governance, but its effectiveness depends on adoption by governments and industry. For the U.S., the voluntary approach aligns with traditional preferences for industry self-governance, whereas for the EU, it fits the future role of the European AI Office in global AI governance.

Despite these international efforts, the U.S. and EU have not established a joint regulatory framework for AI. Instead, the EU has passed pioneering legislation while U.S. regulation focuses on dual-use risks and industry reporting. Both jurisdictions use compute thresholds to identify high-risk models, but their approaches differ in scope and emphasis.