The United States and European Union account for nearly half of global research investment in public interest AI, but they're rarely working together on it. Now, a growing movement to align on shared datasets and privacy standards could unlock breakthroughs in climate forecasting, emergency response, and disease diagnosis that neither region could achieve alone. The key: treating data as a public resource, not a competitive advantage.

What Is Public Interest AI, and Why Does It Matter for Climate?

Public interest AI refers to artificial intelligence systems designed to support long-term survival and well-being of the public, rather than commercial profit or military advantage. Unlike the AI powering chatbots or recommendation algorithms, these systems tackle societal challenges. Examples include improving cancer diagnostics, predicting extreme weather events, enabling sustainable farming, and coordinating disaster relief and emergency response.

The challenge is that building these systems requires massive amounts of high-quality training data. Right now, the US and EU are duplicating effort, each building separate datasets and AI models. A 2023 administrative arrangement between the two regions identified areas of "shared significance and benefit": extreme weather and climate forecasting, emergency response management, health and medicine improvements, and energy grid and agriculture optimization. But progress has stalled, partly because of transatlantic tensions over data privacy and AI regulation.

How Can the US and EU Share Data Without Violating Privacy Laws?

Privacy concerns are real. Healthcare data, in particular, contains sensitive information that can't simply be shipped across the Atlantic. But researchers have developed techniques that allow collaboration without centralizing raw data. The most promising approach is called federated learning, a method where AI models are trained locally on each institution's own data, and only the model updates (not the patient records themselves) are sent to a central server.

A French-American company called Owkin demonstrates how this works in practice. Participating hospitals train models on their own patient data and send only the model updates to a central server, which aggregates them into a global model that's then redistributed to improve each hospital's local version. This way, sensitive information never leaves the hospital, but the collective intelligence benefits everyone.

Privacy-enhancing technologies (PETs) add another layer of protection, though experts note the approach isn't foolproof. Re-identification of data is possible through linkage attacks or inference from behavioral patterns, and anonymity standards differ across jurisdictions. Still, with robust risk management protocols, federated learning and PETs offer a practical path for collaboration among actors that might not otherwise trust each other with sensitive data sharing.

What Public Datasets Already Exist?

Both the US and EU have established frameworks for sharing data responsibly. The examples below show how governments are balancing openness with protection:

• US National Institutes of Health All of Us Dataset: Provides researchers with controlled access to de-identified health data from more than one million Americans, with clear governance safeguards and accountability mechanisms.
• Eurostat Microdata: Grants restricted access to researchers under conditions that rigorously protect the anonymity of individuals and businesses, following EU privacy standards.
• Data.gov: A centralized metadata catalog mandated by the US Open Government Data Act, providing access to datasets published and maintained by federal agencies.
• EU Open Data Directive: Establishes a legal framework for open data, including "high-value datasets" spanning geospatial, earth observation, environment, meteorology, statistics, companies, and mobility data.
• Copernicus Earth Observation Program: The EU's space program offers a dedicated "Data Space Ecosystem" with open datasets to support environmental monitoring, disaster response, and climate analysis.
• Massachusetts AI Hub Data Commons Collaborative: At the subnational level, this directory of datasets enables AI innovation in life sciences, healthcare, climate tech, and education.

How Can Governments Strengthen Transatlantic AI Cooperation?

The institutional foundations for cooperation already exist. As early as 2016, the EU and US launched the Transatlantic Open Data Partnership, bringing together Eurostat and the US Bureau of Economic Analysis to improve interoperability and access to economic data. The 2023 administrative arrangement on AI and computing built on that foundation, allowing researchers to access more data-rich resources by applying federated learning methods and deepening discussions on privacy-preserving AI for biomedical research.

But momentum has slowed. Experts argue that if a planned US-EU dialogue on digital policy materializes, policymakers must revive the public interest workstreams of the 2023 administrative agreement and 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 positioned to support this effort. Research-focused multilateral forums such as the All Atlantic Ocean Research and Innovation Alliance and the International Rare Diseases Research Consortium also offer venues for cooperation.

Steps to Build Stronger Transatlantic Public Interest AI

• Pool Resources on Shared Datasets: The US and EU should combine their research investment to build datasets that capture geographic and demographic diversity in both regions, enhancing representativeness and strengthening the foundation for public interest AI.
• Set Interoperable Standards: Establish common rules about data purpose, access conditions, and accountability mechanisms so that data is formatted, secured, and quality-controlled to a standard that facilitates cross-border research collaboration.
• Implement Federated Learning and Privacy-Enhancing Technologies: Deploy methods that allow institutions to train AI models locally without centralizing sensitive data, reducing privacy risks while enabling collaboration on healthcare, climate, and other sensitive domains.
• Engage Subnational Governments and Research Forums: Support state and member-state governments with open data programs, and leverage multilateral research forums to advance cooperation beyond federal-level negotiations.

The stakes are high. Climate forecasting, disaster response, and disease diagnosis are problems that don't respect borders. By pooling data and expertise, the US and EU could build AI systems that serve billions of people, not just shareholders. The technical tools exist. What's needed now is political will to treat public interest AI as a shared priority, even as the two regions continue to disagree on broader AI regulation.