Global AI Governance: EU Act, UN Frameworks, and What the World Is Deciding

The EU Artificial Intelligence Act (Regulation 2024/1689) is the world's first comprehensive, legally binding AI governance framework. It was passed by the European Parliament in March 2024, signed in July 2024, and phases to full enforcement in August 2026.

How the risk tiers work

• Unacceptable risk (banned): Social scoring systems, real-time biometric surveillance in public spaces, AI that exploits psychological vulnerabilities
• High risk (regulated): AI used in hiring, credit scoring, medical devices, critical infrastructure, law enforcement, and education. Requires conformity assessment, human oversight, and transparency
• Limited risk (transparency requirements): Chatbots and deepfakes must disclose they are AI-generated
• Minimal risk (no requirements): AI used in spam filters, video games, etc.

Why U.S. companies are affected

Any company deploying AI to EU users — regardless of where the company or its infrastructure is based — must comply. U.S. hyperscalers (Microsoft, Google, Amazon, Meta) are full-enforcement targets. The data centers they are building in Michigan to support their AI products serve EU customers.

The Brussels Effect: Because global companies build one compliance system rather than 50, the EU AI Act effectively sets global standards. U.S. companies are building EU-compliant AI systems even for U.S.-only products because the cost of maintaining two systems is higher than compliance.

Starting with the Bletchley Declaration (2023), allied governments have run an annual AI safety summit process — Bletchley (UK) → Seoul (South Korea) → Paris (France) — producing a layered international framework for frontier AI oversight.

What the Paris 2025 framework requires (voluntary)

• Pre-deployment safety testing for frontier AI models above defined capability thresholds
• Disclosure of known safety risks to governments and the public
• Red-teaming and adversarial testing before public release
• Cooperation with national AI Safety Institutes for incident reporting

U.S. posture shift

The Biden administration signed the original Bletchley Declaration in 2023 and created the U.S. AI Safety Institute (AISI) within NIST. The Trump administration declined to re-endorse the Paris Declaration in February 2025 and has reduced AISI funding, signaling a preference for voluntary industry self-governance over international coordination.

Practical effect: U.S. AI companies still voluntarily follow these standards for European market access and reputational reasons. But the U.S. government is no longer an active voice in writing the next iteration of the framework.

U.S. federal AI policy has undergone a significant directional shift between the Biden and Trump administrations:

The gap between U.S. federal AI policy and allied-country frameworks is now the widest it has been since AI governance became a policy field. U.S. companies are navigating compliance with EU law, voluntary adherence to international safety frameworks, and a domestic policy environment that actively discourages mandatory safety requirements.

The most politically consequential AI governance question for the 2026 midterms is not safety standards — it is jobs. IMF research (World Economic Outlook, April 2025) provides the clearest cross-national picture:

• 40% of global jobs will be significantly affected by AI within a decade — higher share in advanced economies
• Sectors most exposed: Financial services, legal, healthcare administration, logistics, customer service, software development
• Sectors less exposed: Skilled trades, construction, childcare, direct personal care, emergency services
• The split within sectors: AI automates specific tasks, not whole jobs — but the tasks it automates are often the entry-level tasks that create career ladders. Displacement may slow career progression more than eliminate positions outright.

The West Michigan picture

Kent County's economy includes healthcare (Spectrum Health/Corewell), financial services (Mercantile Bank, independent advisors), professional services firms, and logistics warehousing. The IMF's high-exposure sectors are well represented. The same data centers being built near Cascade Township will run the AI systems that automate tasks in these sectors.

The governance question: who decides the pace and terms of that transition? Current U.S. federal policy prioritizes infrastructure buildout. The international frameworks prioritize worker transition support alongside deployment standards. Voters in 2026 are choosing which approach their representatives take.

Key AI governance milestones landing in the 2026–2027 window:

• August 2026: EU AI Act full enforcement. High-risk AI systems require conformity assessments. AI content generated for political campaigns requires disclosure.
• Fall 2026: U.S. midterm elections. Congressional control will determine whether federal AI legislation (Data Center Transparency Act, HR 4471; AI Act equivalent legislation) can advance.
• 2027: EU AI Act's general-purpose AI model provisions take effect. This directly affects large language models like GPT-4 and Claude — requiring capability disclosure and safety evaluations.
• Ongoing: Michigan PSC Case U-22103 — how grid upgrade costs from data center load are allocated to ratepayers. EGLE water withdrawal guidance for AI cooling systems. HB 5814 (model local data center ordinance).

Where to follow developments:

• EU AI Act: artificialintelligenceact.eu
• U.S. NIST AI Risk Management Framework: nist.gov/artificial-intelligence
• Michigan HB 5814: legislature.mi.gov → search "HB 5814"