The Regulatory Divergence Problem
Something genuinely consequential is unfolding across the Atlantic, and it’s getting almost no attention. The EU’s AI Act obligations for high-risk systems entered full force in August 2025, backed by fines up to 35 million euros or 7% of global annual turnover, whichever is higher. At the same time, the Trump administration rescinded Biden-era AI safety directives through executive order in January 2025, explicitly prioritizing deployment speed over precautionary regulation. These moves represent fundamentally incompatible philosophies about how democracies should manage transformative technology, and they’re creating real operational friction for companies caught between them.
The stakes are real precisely because we’ve been here before. When the EU implemented GDPR in 2018, the international technology sector absorbed significant compliance costs but eventually developed operational structures to manage dual-standard requirements. AI governance is a harder problem. Data protection is a mature regulatory domain with well-established legal precedent. AI governance is still experimental. Different regulatory frameworks aren’t merely imposing different standards on the same underlying technology—they’re built on different assumptions about what the technology fundamentally is and what governments can or should do about it.
Understanding this divergence means getting past the lazy framing that the EU prefers “regulation” while America prefers “innovation.” That obscures more than it clarifies. The actual disagreement runs deeper, touching questions about burden of proof, the nature of technological risk, who bears responsibility for harm, and whether government should pre-emptively restrict capabilities or respond after problems emerge. These aren’t questions with empirically correct answers. They’re political choices embedded in different democratic traditions.
The Compliance Architecture Clash
Look at what actually happened in the third quarter of 2025. OpenAI, Google DeepMind, and Meta all submitted formal compliance documentation to the EU AI Office in response to the Act’s high-risk system requirements. The same companies, during the same period, were lobbying the U.S. Commerce Department to resist adopting equivalent standards. This isn’t hypocrisy—it’s rational corporate behavior responding to different legal environments. But the pattern reveals something structurally important about how divergent regulation works in practice.
The EU framework requires developers of high-risk AI systems to conduct documented risk assessments before deployment, maintain detailed records throughout the system’s operational life, establish human oversight mechanisms, and submit to periodic audits by accredited third parties. These requirements impose genuine operational costs. A Stanford HAI policy brief released in October 2025 estimated that regulatory divergence between EU and U.S. frameworks generates roughly 4.2 billion dollars annually in compliance expenses for multinational AI developers. That figure includes duplicate compliance infrastructure, separate documentation systems, and the substantial legal expertise required to navigate both regimes.
What makes this particularly significant isn’t just the cost—it’s the incentive structure it creates. When compliance is expensive and fragmented, larger companies with robust legal and compliance divisions absorb those costs far more easily than smaller competitors. The effect, counterintuitively, may be to entrench market concentration rather than encourage competition. A startup operating in both markets faces proportionally higher compliance burdens than Microsoft or Google. Neither regulatory regime explicitly addresses this market-distortion risk, which strikes me as a serious oversight.
The Three-Way Fragmentation
The situation gets messier when you factor in the third major regulatory power. China’s Cyberspace Administration finalized its second round of generative AI regulations in mid-2025, establishing a third distinct governance framework operating according to its own logic and priorities. These aren’t marginal players. Between them, the EU, United States, and China represent the overwhelming majority of global AI development capacity and nearly all the cutting-edge research institutions capable of producing frontier AI systems.
The OECD characterized this moment in 2025 as “the most consequential splintering of technology governance norms since GDPR.” That framing captures something real, even if it treats fragmentation as settled fact rather than a problem to solve. With GDPR, there was at least a unifying principle—data subjects have rights that companies must respect. The current AI governance landscape lacks even that degree of conceptual coherence. The EU’s framework emphasizes pre-deployment risk management and human oversight. The American approach emphasizes post-deployment transparency and user redress. The Chinese framework emphasizes security and social stability. These aren’t different points along a shared axis. They’re genuinely different conceptions of what AI governance is supposed to accomplish.
The practical consequence is that developers must now design, deploy, and operate AI systems according to potentially contradictory requirements. A content moderation system acceptable under EU risk assessment requirements might violate American First Amendment principles or Chinese sovereignty requirements. These aren’t trivial incompatibilities. They require either abandoning certain markets, maintaining separate product versions, or making substantive compromises to core system design.
What the Evidence Actually Shows
We should be careful about overconfident predictions regarding which regulatory model will “win” or whether convergence remains possible. History suggests that network effects and market power typically drive technical standards toward a dominant model, but governance standards operate differently. The EU AI Act is already law with enforcement mechanisms in place. The American regulatory framework, by contrast, remains scattered across multiple agencies with no comprehensive statutory foundation. The Chinese system consists of state directives issued through administrative channels. These aren’t equivalent structures operating on equivalent timelines.
What the Stanford HAI AI Index and policy briefs actually document isn’t a prediction about which model will dominate—it’s a measurement of current compliance costs and market impacts. That data shows real economic friction is already here. Companies report longer time-to-market for AI products in certain jurisdictions, strategic decisions to prioritize specific geographic markets, and increased legal and compliance staffing. These are observable facts, not speculative projections.
The evidence also suggests that regulatory convergence isn’t currently happening, despite the stated preference for harmonization from multiple international organizations. Neither the EU nor the U.S. has shown willingness to substantially modify its approach. China’s framework keeps evolving according to its own strategic priorities. If convergence were actually happening, we’d expect harmonization proposals to be gaining serious traction at the policy level. The absence of that movement suggests each jurisdiction views its approach as reflecting legitimate national values, not as a temporary position awaiting alignment.
The Deeper Question About Democratic Governance
What ultimately matters about this regulatory divergence isn’t just the operational complexity it creates for technology companies, though that’s real. What matters is what it reveals about how democracies approach collective decisions about transformative technology in an interconnected world. The EU chose to move first with comprehensive legislation built on precautionary principles. The U.S. proceeded through distributed sectoral regulation coupled with speed prioritization. Both choices reflected legitimate democratic values and constituency preferences. Neither was objectively wrong.
The problem is that both choices assume regulatory authority at the national or supranational level can meaningfully constrain technology that operates globally. That assumption may not hold. If AI development concentrates in jurisdictions with lighter regulatory frameworks, and if the systems produced there become de facto global standards through network effects and market penetration, then strict regulation elsewhere may constrain only domestic companies and innovation—not the technology itself. This is the genuine dilemma facing policymakers, and it has no purely technical solution.
The fragmentation we’re watching in 2025 and 2026 will likely reshape how we think about technology governance more broadly. If regulatory divergence produces inferior outcomes for everyone—higher compliance costs, market distortion, persistent governance gaps—it may eventually push toward either genuine harmonization efforts or toward accepting that some technological domains simply aren’t amenable to meaningful regulation through existing state structures. Neither option is comfortable, but both are at least intellectually honest responses to what the evidence is telling us. What interpretations of this moment do you find most persuasive? What governance models do you think better balance innovation, safety, and democratic legitimacy?