Where Transatlantic AI Cooperation Can Still Work

Debate on transatlantic technology relations has often overlooked areas of potential cooperation and opportunities for shared learning and benefits for the American and European publics. Unlike AI developed for commercial or military use, public interest AI advances societal goals such as education and public health, areas in which transatlantic cooperation can make a meaningful difference. Public interest AI is created on a value chain intentionally built to serve society at every stage, but any progress on this goal depends on upstream choices about the data gathered, from whom and under what conditions it is collected, and the safeguards in place during the process. Data and its curation into high-quality public datasets, after all, play a central role in the development of AI systems.

Public Interest AI is a debated term in academic and policy circles but generally refers to AI systems designed and deployed with the aim of supporting the long-term survival and well-being of the public. It encompasses applications for a range of uses such as improving cancer diagnostics, predicting extreme weather events, enabling sustainable farming, and coordinating disaster relief and emergency response. 

Transatlantic tension and division on technology regulation and data privacy, however, create a real risk of undermining critical public interest AI research. The United States and the EU account for nearly half of global research and development investment in this area but are, fortunately, well placed to manage this risk. By pooling resources and setting interoperable standards, they could build datasets that capture the geographic and demographic diversity in both regions, enhance their representativeness, and strengthen the foundation for public interest AI.

A Link in the AI Value Chain

Public datasets are curated collections of data made available by public-sector bodies. As a foundational building block for public interest AI, they provide the training data, benchmarks, and shared resources on which these systems depend. When governed by clear rules about purpose, access conditions, and accountability mechanisms, public datasets ensure that data is formatted, secured, and quality-controlled to a standard that facilitates cross-border research collaboration.

Public datasets established by US and EU government bodies illustrate this approach by combining governance safeguards with controlled access. The US National Institutes of Health’s All of Us dataset, for example, provides researchers with controlled access to de-identified health data from more than 1 million Americans, while Eurostat’s Microdata grants restricted access to researchers under conditions that rigorously protect the anonymity of individuals and businesses.

Both jurisdictions also maintain established open data frameworks for accessing public datasets. In the United States, Data.gov, mandated by the Open Government Data Act, serves as a centralized metadata catalogue, providing access to datasets published and maintained by federal agencies. Similarly, the EU’s Open Data Directive establishes a legal framework for open data that includes the re-use of “high-value datasets”. These are intended to generate societal and economic benefits to enable AI and data-driven innovation. They span six thematic areas: “geospatial, earth observation and environment, meteorology, statistics, companies, and mobility”. Copernicus, the Earth observation component of the EU’s space program, has a dedicated “Data Space Ecosystem” that offers access to open datasets to support environmental monitoring, disaster response, and climate analysis. At the subnational level, the Massachusetts AI Hub’s Data Commons Collaborative offers a directory of datasets to enable AI innovation in sectors such as life sciences, healthcare, climate tech, and education.

In highly sensitive areas in which cross-border data centralization raises legal and privacy concerns, such as healthcare, the use of public datasets in combination with methods such as federated learning (FL) could allow models to be trained locally without raw data ever being centralized. Healthcare institutions, for example, could collaborate on joint analysis and model training without sharing sensitive patient data. Owkin, a French-American AI company, uses a federated software for biomedical research, through which participating hospitals can train models on their own data and can send only model updates—not patient data—to a central server. The server aggregates updates into a global model, which is then redistributed to the hospitals to retain and improve their models.

Researchers have shown that Privacy-Enhancing Technologies (PETs) can also enhance privacy and security of data and communications within FL frameworks, although this approach is not straightforward. Re-identification of data is possible through linkage attacks, combining datasets, or inference from behavioral patterns. In addition, anonymity standards differ across jurisdictions and firms. With robust risk management and mitigation protocols, however, FL approaches and PETs offer a promising path for collaboration among actors that might not otherwise trust each other with sensitive data sharing.

A Base To Build On

Examples of transatlantic cooperation on data and public interest AI provide a meaningful foundation for further collaboration. As early as 2016, the EU–US Transatlantic Open Data Partnership brought together Eurostat and the US Bureau of Economic Analysis to improve interoperability and access to economic data. In 2023, the EU and the United States signed an administrative arrangement on AI and computing to advance AI research in areas of “shared significance and benefit”: extreme weather and climate forecasting, emergency response management, health and medicine improvements, and energy grid and agriculture optimization. The agreement allows researchers to access more data-rich resources by applying methods such as FL and deepens discussions on privacy‑preserving AI for biomedical research.

Such an initiative demonstrates that institutional foundations for transatlantic cooperation already exist. The next step is deliberate, continuous progress on public interest AI despite challenges in transatlantic technology ties. Should the planned US-EU dialogue on digital policy materialize, policymakers must consider reviving the public interest workstreams of the 2023 administrative agreementand dedicate resources to developing interoperable datasets for public interest AI. Beyond the federal level, US state governments with mature open data programs and their European member-state counterparts are well placed to support this. Research-focusedmultilateral forums such as the All Atlantic Ocean Research and Innovation Alliance and the International Rare Diseases Research Consortium also offer venues for cooperation. Together, all these measures offer a practical way to leverage collective strengths in public interest AI even when broader AI policy alignment remains elusive.

The views expressed herein are those solely of the author(s). GMF as an institution does not take positions.