The Regulatory Divergence We’re Actually Watching
We live in a moment when the world’s two largest economic blocs are not merely disagreeing about how to regulate artificial intelligence. They are disagreeing about the fundamental premise of regulation itself. This distinction matters more than headlines about “strict EU rules” versus “light-touch American approach” typically suggest. The European Union and the United States have entered a period where their governance frameworks are becoming actively incompatible, not just differently stringent. Understanding why requires moving beyond the surface-level narrative and examining what each system is actually trying to accomplish.
The EU AI Act’s enforcement phase commenced on August 2, 2025, marking the moment when this regulatory framework stopped being aspirational and became binding law. Developers and companies deploying high-risk AI systems must now conduct conformity assessments, maintain mandatory audit logs, and implement human oversight mechanisms. The financial teeth are real: penalties reach up to 30 million euros or 6 percent of global turnover, whichever is larger. These are not theoretical constraints. For companies operating at scale, this represents a material business calculation that shapes product development timelines, staffing decisions, and deployment strategies.
Simultaneously, the Trump administration moved swiftly to dismantle the Biden-era executive order on AI safety signed in October 2023. Within the first weeks of this administration, Executive Order 14110 was revoked, eliminating the mandatory safety reporting requirements that had obligated frontier AI developers to share results of internal testing with federal authorities. This was not a modest recalibration. It was a strategic reversal signaling a fundamentally different view about whether the federal government should have visibility into advanced AI safety testing before systems enter commercial deployment.
The Enforcement Question: Where Power Actually Resides
One of the hardest things to convey to people not steeped in regulatory policy is that enforcement mechanism design often matters more than the rules themselves. A regulation without teeth is rhetoric; a regulation with enforcement capacity is governance. The EU has built actual enforcement infrastructure. Member states have appointed competent authorities. The European Commission has allocated staff and budget. This is not costume jewelry regulation. It is a system designed to function.
The Biden executive order, by contrast, operated through what we might call “regulatory signaling.” It carried no criminal penalties or statutory fines. Its power derived from presidential direction within the executive branch and the understanding that companies would comply with federal guidance to maintain their relationships with government agencies. This proved fragile. When administrations change, executive order governance can be dismantled just as readily as it was constructed.
What happened to the National Institute of Standards and Technology framework illustrates this further. NIST’s AI Risk Management Framework was not formally rescinded. Instead, a March 2025 Office of Management and Budget circular revision eliminated the interagency mandate requiring federal agencies to incorporate it into procurement and policy decisions. The framework still exists. You can access it. But it no longer carries the institutional weight that transforms best practices into federal requirements. This is how regulatory rollback often works in the American system: not through dramatic proclamation but through administrative shuffling that quietly removes enforcement leverage.
The Market Concentration Problem and Transatlantic Friction
Here sits one of the most consequential facts in contemporary technology policy that doesn’t get nearly enough political attention: the concentrated geography of AI model development. The Stanford HAI research organization published its 2025 AI Index Report documenting that 19 of the 25 largest AI model releases in 2024 originated from U.S.-based companies. This is not merely a technological observation. It is a regulatory jurisdiction question with profound implications. When the overwhelming majority of frontier AI systems are developed by American companies, regulatory choices made in Washington ripple through the entire ecosystem.
The EU faces a genuine structural problem. It has written stringent rules that apply to high-risk systems, yet most of the companies developing those systems operate from American soil under a different regulatory regime. This creates what policy analysts call regulatory arbitrage opportunity. A U.S. company can develop an AI system under permissive American guidelines, then deploy that same system to European users who are theoretically subject to stricter EU requirements. The company must technically comply with EU rules in the European market, but the underlying development process operated under minimal federal scrutiny.
This tension has become so acute that the EU-U.S. Trade and Technology Council’s AI subcommittee suspended formal coordination sessions in June 2025. These suspensions are not casual scheduling conflicts. They represent an admission that the two systems have reached irreconcilable positions on fundamental questions like mandatory pre-deployment testing standards. The Council previously functioned as a mechanism to find common ground and coordinate approaches. Its inability to do so suggests the divergence has moved beyond technical disagreement into genuine policy opposition.
Why This Matters Beyond Silicon Valley
The practical implications extend far beyond technology companies and their compliance departments. Consider what regulatory fragmentation means for innovation itself. A company deciding whether to develop a particular AI capability must now calculate which markets it can profitably serve. If the EU enforcement regime proves sufficiently stringent that certain applications become economically unviable in European markets, companies face a choice: absorb the compliance costs or forgo the European market. Either decision shapes what technologies actually get built and tested.
There are also geopolitical layers worth acknowledging. The EU’s regulatory approach reflects a genuine political commitment to constraining powerful technologies through law, philosophically rooted in European thinking about individual rights and the state’s responsibility to protect citizens from concentrated private power. The American approach reflects a different philosophical inheritance: skepticism toward centralized regulatory authority and confidence that market competition and liability law create sufficient incentives for responsible behavior. Neither approach is obviously wrong. But they cannot easily coexist in a globally integrated market.
For citizens trying to understand what this means for their own lives, the point is this: you likely use AI systems developed by American companies operating under American regulatory frameworks, even if you live in the European Union. How stringently those systems were tested before deployment, whether developers maintained detailed logs of their performance, and whether human oversight mechanisms are actually functional now depends partly on which regulatory regime your national government can enforce jurisdiction over. That is a genuinely complicated question without a neat resolution.
What We Should Be Watching
Three developments in the months ahead seem worth tracking closely. First, watch whether the EU actually enforces its high-risk system requirements against major technology companies. Regulatory credibility depends on demonstrable enforcement. Second, monitor whether the Trump administration faces pressure to articulate an affirmative AI governance approach, or whether the strategy remains purely deregulatory. Absence of rules is not the same as deliberate governance. Third, pay attention to whether American companies begin fragmenting their development practices, maintaining separate systems optimized for different regulatory markets, or whether they absorb compliance costs and implement EU-style practices globally.
The deeper question underlying all of this is whether a technology as consequential as artificial intelligence can exist in a world of radically incompatible governance frameworks. Evidence from other domains suggests that eventually some form of alignment becomes necessary, though the path there is rarely direct. For now, we are in a period of genuine uncertainty where two substantial economic powers have chosen substantially different regulatory architectures.
You can review the EU’s detailed implementation guidance through the European Commission EU AI Act Implementation Portal, which documents the specific requirements for conformity assessments and reporting obligations. For a comprehensive data-driven perspective on AI development patterns globally, the Stanford HAI Artificial Intelligence Index Report 2025 provides detailed analysis of where models are actually being created and what that means for regulatory jurisdiction questions.
AI governance will likely define significant dimensions of technology policy for the next decade. What questions do you see as most critical in this transatlantic regulatory divergence? I would welcome hearing your perspective on which developments you are watching most closely.