FDA builds AI-ready data platform on Databricks for government to secure regulatory work

The FDA migrated more than 5,000 users and 8,000 jobs to a governed data platform on AWS GovCloud with zero downtime, while the agency regulates 20 cents of every dollar of U.S. consumer spending.

Categorized in: AI News Government
Published on: Sep 02, 2026
FDA builds AI-ready data platform on Databricks for government to secure regulatory work

The U.S. Food and Drug Administration built and migrated its entire enterprise data platform to a new governed foundation on Databricks for Government running on AWS GovCloud, moving more than 5,000 users and 8,000 jobs with zero downtime. The overhaul matters because the FDA regulates 20 cents of every dollar of U.S. consumer spending, and its decisions touch one in three Americans every day.

The platform, called HALO (Harmonized AI and Lifecycle Operations for Data), consolidates more than 40 data sources across eight centers and 30 programs. It uses Unity Catalog to provide a single governance layer for data and AI assets, so teams can discover, manage, and share trusted data without the operational overhead that plagued the agency's previous fragmented systems.

Three milestones that enabled the shift

The FDA's modernization hinged on three sequential milestones. The first was FedRAMP High authorization sponsorship, which unlocked the regulatory path forward. The second was the migration to AWS GovCloud. The third was adopting Unity Catalog, which gave the agency the governance foundation required for secure, scalable AI workloads.

The migration itself was extensive. Teams refactored more than 1,000 data pipelines and more than 4,000 notebooks. They completed the move while scientists and analysts continued their regulatory work uninterrupted. The FDA described its multi-tenant architecture as an apartment complex model: every center shares the infrastructure but maintains its own secured, governed space with distinct policies and locks.

Measurable operational gains

The results were concrete. SQL query response times for BI workloads improved by more than 30%. Compute costs fell by more than 20%. Time spent on provisioning, permissioning, and data sharing dropped by more than 75%. Operational overhead declined by more than 35%.

User adoption grew from roughly 500 users in 2020 to more than 6,000 today. The agency expects that number to exceed 10,000 by 2028.

"Authorization is an accelerant. Governance is the prerequisite for AI," the FDA team said during their session. "Large-scale modernization can move faster when teams run critical workstreams in parallel instead of waiting for perfection."

Responsible AI in a regulatory environment

The FDA integrated HALO with Elsa, its enterprise AI platform, to support AI use cases already in production and more than 10 in development. One flagship initiative, called MARS, helps reviewers analyze massive volumes of structured and unstructured data from drug and device applications, including clinical trial results, safety data, and labeling. The goal is to get the right information to the right reviewer faster and reduce time spent on data wrangling.

The agency stressed a human-in-the-loop approach. AI does the legwork and augments the expertise of scientists and reviewers but does not replace them. The modernization was not about AI for its own sake. It was about enabling responsible AI under the stringent compliance and auditability requirements of a FedRAMP High environment.

For government agencies pursuing similar paths, the FDA's experience surfaces practical lessons. Stakeholder engagement is continuous because every center has different needs, timelines, and risk tolerances. Security planning must start early, as every architectural decision has downstream implications in a regulated environment. The agency also recommended building for flexibility, adopting wave-based migrations instead of big-bang cutovers, and treating vendor partnerships as operational assets.

Why this matters for government leaders

The FDA's story demonstrates that secure AI in government does not start with models. It starts with the governed data foundation underneath them. When governance is built in from the start, modernization reduces technical debt and creates the conditions to put AI into the hands of scientists, analysts, and decision-makers securely. For public sector leaders evaluating AI for Government, the takeaway is practical: compliance and speed are not trade-offs when the platform architecture treats governance as a first-class requirement, not an afterthought.


Get Daily AI News

Your membership also unlocks:

700+ AI Courses
700+ Certifications
Personalized AI Learning Plan
6500+ AI Tools (no Ads)
Daily AI News by job industry (no Ads)