How Wide is the Atlantic? Convergence and Divergence in US–EU AI Governance

Download this commentary (PDF)

The GW Regulatory Studies Center and the Frankfurt Competence Center for German and Global Regulation co-hosted a virtual conference, Regulatory Crosswinds, on whether the United States and the European Union can stay aligned amid administrative upheaval. The opening panels discussed regulatory reforms broadly in the U.S. and Europe and are summarized in a companion commentary. The second half focused on AI governance and the narratives that Europe regulates, while America innovates, or that Europe guards data, while America builds on it. Remarks by Joseph Dunne, Research Professor at the GW Institute of Public Policy, and Andreas Jungherr, Professor of Political Science at the University of Bamberg and a director of the Bavarian Research Institute for Digital Transformation, took that claim apart. The Atlantic is narrower than it sounds, and the gap worth watching runs elsewhere.

A Common Starting Point, Two Different Paths

Dunne observed that both sides of the Atlantic built their AI policy on three shared concerns: that systems be trustworthy, that they be safe, and that they hold their own in a contest with rivals. Each side stood up a safety body. Each leaned on standards, and the NIST risk-management framework and the European Union (EU) AI Act now rhyme more than they clash. The paths then forked.

Dunne argued that Europe legislates to protect rights; America leaves matters to the market and the courts. Brussels’ industrial policy focus relied on legislation and centralization across the EU. In contrast, the U.S. focused more on soft law approaches. However, individual U.S. states have begun to legislate in ways similar to Europe, despite a Trump executive order to preempt state AI laws and cut funding to states that resist. California’s SB 53 requires transparency from frontier developers. New York’s RAISE Act adds incident reporting. Illinois has passed SB 315, the first law in the country to demand independent third-party safety audits.

Europe treats privacy as a fundamental right, set in Article 8 of the Charter and in the General Data Protection Regulation (GDPR); America treats it as consumer protection, with no stand-alone federal statute and a patchwork of state law. That difference strains the machinery that keeps data moving. The EU–US Data Privacy Framework, the third such deal after two were struck down, rests on American oversight bodies. When the Trump administration removed three members of the Privacy and Civil Liberties Oversight Board in January 2025, a court ordered two of them back, an appeals court stayed that order, and the question now sits with the Supreme Court. The framework stands on an unsteady base.

The European Commission is now focusing on sovereignty. On June 3, it unveiled its Technological Sovereignty Package, four measures built to cut Europe’s dependence on foreign technology. Its Cloud and AI Development Act sorts providers into four sovereignty levels; American firms may compete in the lower two, while the top tiers ask for a European hand.

What the Public Actually Thinks

One might expect different approaches to reflect different citizen preferences across jurisdictions, but Jungher’s research challenges that notion. His institute surveyed nine countries and 9,000 people on the hardest cases: AI in the military and AI in policing. Contrary to the conventional wisdom that the German public is more fearful of AI than Americans, the data show broad similarity. Both populations lean toward acceptance, rating most uses above the midpoint of a seven-point scale. The clearest resistance comes over fully autonomous lethal force, weapons that select and kill without human control, and even there the divide is not German against American but lethal against everything else, as both countries’ populations appear similarly cautious.

Germany is not the most enthusiastic population, but neither is it the least. On the broad measure of how central people think technology is to national success, it lands just behind the leaders, Taiwan, China, and the United States, and ahead of the other European countries surveyed. Jungherr tied that gap not to any fear of AI but to a weaker German conviction that technology drives wealth, influence, and security. Across both jurisdictions, what best predicts support is not the specific use but how a person views AI in general; perceived benefit, not nationality, is the strongest driver. The implication he drew for policy is to make the broad case for technology’s contribution to national capacity, rather than defend one application at a time or run narrow literacy campaigns.

Dependence and the Brussels Effect

During the audience discussion, questioners asked whether Europe’s turn toward sovereignty could actually work. Jungherr was doubtful. Europe’s economy and security run on American technology, and Washington has shown it can leverage that dependence. But drawing on the idea of weaponized interdependence, he argued that coercion through dependence is self-limiting: it accelerates the decoupling it seeks to prevent, as targeted states (e.g., Europe)  move to diversify away from the source of pressure (e.g., the U.S.). He offered managed interdependence as the honest aim, since the United States leans on European talent as surely as Europe leans on American infrastructure. Dunne agreed the European goal is to reduce dependence, but read the sovereignty package as buying time rather than building walls; its levels invite American partners into the lower tiers, and Europe, he noted, still wants to partner. He pointed to the EU’s Digital Omnibus, an effort to simplify its own rules, as a sign that Brussels is adapting, even if the effort sits uneasily with careful regulatory analysis.

Pressed on whether Europe still sets the world’s rules, the two parted again. Dunne recalled that the AI Act was conceived to repeat the “Brussels effect,” the pull by which EU rules become global defaults, but judged that influence unlikely to arrive this time. The effect, Jungherr argued, was always more hope than reality, joking that it was “Brussels fan fiction.” The GDPR’s heavy compliance costs, he added, offer a warning rather than a model for other countries.

The Gap That Matters

Read together, the two talks reach two conclusions. The first suggests more convergence between jurisdictions than is commonly appreciated. Dunne’s map of institutions and Jungherr’s survey of public opinion run in the same direction: the U.S. and the EU began from shared concerns; they still trade standards and borrow instruments, and diverge far less than conventional wisdom suggests, even as Europe’s sovereignty turn marks a real fork. The second conclusion is that policy differences may stem less from different public perceptions than different drivers. The transatlantic gap over what the public will accept is small; the gap that matters divides elites from the rest of society. As Jungherr cautioned, elite and media discourse can overstate public resistance, especially early in a technology’s life. He also noted that Washington’s enthusiasm is industry-led, and its system stays open to disruption, and is comfortable with new firms displacing incumbents. In contrast, officials in Brussels and Berlin are more comfortable regulating to shield established European industries.