Atlantic Council Data governance for AI and Healthcare: From Washington to Beijing

On Tuesday, September 8, the Civitas One Team attended the Atlantic Council’s Streamed Public Discussion: Data governance for AI and Healthcare: From Washington to Beijing. The full discussion is available here.

Speakers of the discussion are:

• Ranjit Kumble, Vice President, Trusted AI Pfizer

• Stephen Moon, Global Public Sector Chief Technology Officer, Snowflake

• Daria Bahrami, Head of Policy, Dreadnode

• Gil Alterovitz, Associate Professor, Harvard Medical School; President, Presidential Innovation Fellows Foundation

Moderated by:

• Graham Brookie, Vice President, Atlantic Council Technology Programs

• Trey Herr, Assistant Professor, School of International Service, American University; Nonresident Senior Fellow, Cyber Statecraft Initiative, Atlantic Council

The core takeaway was unambiguous: while AI relies on borderless data flows, global data governance is pulling in the opposite direction toward national sovereignty, strict localization mandates, and fragmented regulatory regime.

Fireside Chat: Building AI for Healthcare in a Fragmented Data Environment

The opening session featured a conversation between Ranjit Kumble and Graham Brookie, focused on how global biopharmaceutical leaders balance rapid AI innovation with strict regulatory compliance across international markets.

AI models deployed in drug discovery and clinical trial optimization require access to massive biological, genomic, and demographic datasets to achieve statistical accuracy. However, biopharmaceutical companies must navigate an increasingly fragmented landscape where national boundaries dictate data movement. Severe data localization mandates in China, controls on consumer protections, and evolving sector-specific rules in the U.S. create severe legal friction when trying to pool patient records into centralized repositories for global model training.

Pfizer, specifically, has embedded “Trusted AI” governance directly into its early-stage R&D workflows rather than treating regulatory alignment as a later legal checkpoint. This requires teams to maintain strict data provenance tracking, establish model explanations and conduct algorithmic bias audits throughout the entire software lifecycle. The panelist emphasized that deployable healthcare AI relies heavily on reliable compute capacity, cloud data storage, and resilient energy grids capable of supporting intensive model training workloads under sovereign data laws.

Panel Discussion: Navigation Challenges

Moderated by Trey Herr, the panel discussion brought together Kumble, Stephen Moon, Daria Bahrami, and Gil Alterovitz to examine how technological architecture can bridge regulatory divides.

The panelists compared the three primary geopolitical regulatory models governing healthcare data:

Region	Regulatory Ideology	Impacts on Health AI
United States	Market-led innovation with sector-specific rules, including HIPAA and FDA guidance	Agile commercial deployment, but fragmented state-level privacy rules and growing national security scrutiny over bulk data
European Union	Preemptive rights-based regulation, including the EU AI Act	High consumer trust and clear risk classification, but significant administrative overhead for high-risk medical AI tools
China	State sovereignty and data security	Access to massive domestic clinical datasets, but mandatory data localization and strict limits on outbound cross-border data transfers

To overcome cross border legal barriers, enterprise cloud providers and health systems are turning to privacy-enhancing technologies (PETs) like federated learning. By distributing algorithms directly to local data centers or hospital servers, federated architecture allows AI models to train on encrypted local records without physically transferring raw patient files across national boundaries. Only anonymized model parameter updates are sent back to a central server, allowing multi-site international clinical trials to proceed while remaining fully complaint with regional data sovereignty mandates.

However, moving advanced AI into frontline clinical environments introduces unique security vulnerabilities that extend far beyond standard compliance checklists. Diagnostic and predictive algorithms face threats, including data poisoning aimed at corrupting clinical predictions, Panelists stressed that securing AI in healthcare required continuous red-teaming, rigorous model validation, and proactive threat modeling prior to deployment.

Audience Q&A: Implementation, Patient and Consumer Trust

Maintaining public trust requires clear guardrails around how patient information is utilized for commercial AI development. Panelists highlighted synthetic data generation and advanced anonymization protocols as vital mechanisms for unlocking research value. Synthetic datasets mirror the real clinical populations without exposing individual patient records, allowing researchers to conduct initial hypothesis testing and model validation in secure environments without compromising personal privacy.  

Restricting training data within strict national boundaries poses a significant threat to health equity by introducing localized algorithmic bias. AI models trained exclusively on homogenous regional datasets often perform poorly or generate inaccurate clinical predictions when applied to broader, racially diverse populations. To prevent clinical bias, international research networks must preserve secure channels for cross-border data collaboration or leverage training that reflect global genetic and demographic diversity.

Finally, the panel examined how smaller healthcare institutions, research centers, early-stage biotech startups can manage the escalating cost of compliance across sovereign markets. As Washington and Beijing refine their respective regulatory regimes, heavy administrative and infrastructure demands risk favoring deeply capitalized technology and pharmaceutical incumbents. The panelists concluded that international standards organization must establish open baselines to prevent market consolidation and ensure broader access to life-saving AI innovations.

Sources:

• Atlantic Council, "Data Governance for AI and Healthcare: From Washington to Beijing," September 8, 2026. Link