Modernizing a federal information platform is a bit like steering an plane provider whereas rebuilding the engine mid-ocean: the mission can’t cease, even because the underlying techniques are being remodeled. That’s the problem the U.S. Meals and Drug Administration (FDA) described in its Knowledge + AI Summit session on How the FDA is scaling safe AI with Databricks for Authorities.
For the FDA, the stakes are unusually excessive. One in three People is touched by an FDA resolution day-after-day, it regulates 20 cents of each greenback of U.S. client spending, and its greater than 16,000 workers work throughout 200 workplaces and labs protecting greater than 300 product classes. In that atmosphere, information isn’t a back-office perform; it’s a public well being crucial.
That’s the reason the dialog round AI in authorities is shifting. The query is now not whether or not companies ought to use AI, however how they’ll operationalize it securely in environments that demand robust governance, auditability, and compliance from day one. This weblog outlines the core constructing blocks required to help safe, mission-critical AI workloads in FedRAMP and IL5 environments: infrastructure-as-code safety patterns, ruled mannequin entry, lineage and auditability, and the networking configurations wanted to attach Unity Catalog to the appropriate information sources.
Why safe AI in authorities begins with platform readiness
Authorities companies usually function underneath among the most stringent necessities on this planet. Databricks on AWS GovCloud is designed for such regulated workloads that course of and analyze export-controlled information (ITAR/EAR), information that’s regulated underneath FedRAMP Excessive, DoD IL5, or processes different delicate info like healthcare data and federal monetary techniques which can’t be dealt with in an ordinary cloud atmosphere.
However compliance alone isn’t the story. The larger level is velocity to worth. As a part of this session, Databricks highlighted a safety reference structure and Terraform-based deployment mannequin designed to assist prospects get up a hardened, production-ready, audit-ready atmosphere shortly, with controls corresponding to PrivateLink, customer-managed keys, and the compliance safety profile baked in.
The FDA’s problem: modernize with out disrupting the mission
The FDA’s modernization journey started with a well-known drawback: siloed information, duplicated effort, inconsistent pipelines, and an excessive amount of operational overhead unfold throughout a number of facilities and packages. Analysts and scientists have been spending an excessive amount of time discovering and reconciling information as a substitute of deriving perception from it. On the identical time, the hole between the company’s scientific and regulatory mission and what its infrastructure may help was widening.
The company’s response was to construct HALO (Harmonized AI and Lifecycle Operations for Knowledge), its enterprise information platform, as a safe, ruled, AI-ready basis. After beginning its journey with Databricks in 2020 with the modernization of a legacy atmosphere, the FDA’s transfer to Databricks on AWS GovCloud in 2025 unlocked the advantages of Unity Catalog and extra capabilities within the Databricks atmosphere. Unity Catalog gave FDA a single, open governance layer for information and AI to allow them to securely uncover, handle, and share trusted belongings throughout areas, codecs, and instruments with much less complexity and decrease value, whereas bettering the time-to-insights.
The structure the FDA adopted is multi-tenant: a number of regulatory facilities share a typical platform basis whereas sustaining their very own secured, ruled areas. Within the session, FDA described it as an residence complicated mannequin, the place everybody shares the infrastructure however every tenant has its personal lock and insurance policies. With Unity Catalog, it turned a lot simpler to share information throughout facilities with out shedding governance.
Three milestones that modified the trajectory
In FDA’s journey, three milestones outlined the transformation. The primary was FedRAMP Excessive authorization sponsorship, which was the prerequisite that unlocked every thing else. The second was the AWS GovCloud migration. The third was Unity Catalog, which gave the company the governance basis it wanted for safe, scalable AI.
The size behind these milestones is what makes the story particularly compelling. The FDA migrated greater than 5,000 customers and greater than 8,000 jobs and pipelines with zero downtime, permitting scientists and analysts to proceed their work uninterrupted whereas the company rebuilt the inspiration beneath them. It additionally refactored greater than 1,000 information pipelines and greater than 4,000 notebooks as a part of the migration to Unity Catalog.
The FDA is utilizing Databricks serverless compute, mannequin serving, Unity Catalog, and Genie to modernize lots of of legacy dashboards into interactive, AI-powered experiences, eliminating technical debt and accelerating mission affect.
The Impression – What modernization delivered in follow
For the FDA, this was not simply an infrastructure cleanup. It was a measurable working shift. The company stated that, in only a few months, it onboarded eight facilities and 30 packages onto its enterprise information platform and consolidated greater than 40 information sources spanning software and submission techniques. That consolidation improved collaboration throughout the company, elevated transparency, and strengthened its safety posture.
It additionally produced quantifiable operational positive aspects. In accordance with the FDA, SQL warehouses improved question response occasions by greater than 30% for BI workloads, compute prices fell by greater than 20%, time spent on provisioning, permissioning, and information sharing dropped by greater than 75%, and operational overhead declined by greater than 35%.
Consumer adoption tells the identical story. The FDA began with roughly 500 customers in 2020, has now grown to greater than 6,000 customers on the platform, and expects that quantity to exceed 10,000 by 2028.
Enabling accountable AI for regulatory work
One of the vital vital themes that’s value noting right here is that the FDA’s modernization program was not about AI for AI’s sake. It was about enabling accountable AI in a federal regulatory atmosphere. The company has built-in Halo with Elsa, its enterprise AI platform, to help AI innovation at scale, with AI use instances already in manufacturing, in pilot, and greater than 10 in flight.
A flagship instance is MARS, the FDA’s initiative for modernizing and accelerating regulatory submissions. MARS makes use of Databricks built-in with Elsa to assist reviewers analyze huge volumes of structured and unstructured info related to drug and gadget purposes, together with medical trial outcomes, security information, and labeling. The aim is to get the appropriate info to the appropriate reviewer sooner, cut back time spent on information wrangling, and help better-informed regulatory choices.
Simply as vital, the FDA emphasised a human-in-the-loop method: AI does the legwork and augments the experience of scientists and reviewers, however doesn’t change them.
Classes for different companies scaling safe AI
The FDA’s journey surfaces a number of classes that resonate throughout the general public sector. First, stakeholder engagement isn’t a one-time occasion; it’s a steady dedication as a result of each heart has completely different wants, timelines, and threat tolerances. Second, safety planning has to begin early as a result of, in a FedRAMP Excessive atmosphere, each architectural resolution has downstream implications.
The FDA additionally pressured the significance of constructing for flexibility as a substitute of hard-coding configurations, adopting wave-based migrations as a substitute of big-bang cutovers, and treating a powerful vendor partnership as an operational asset quite than a procurement checkbox.
The closing message from the session was particularly clear. Authorization is an accelerant. Governance is the prerequisite for AI. And huge-scale modernization can transfer sooner when groups run important workstreams in parallel as a substitute of ready for perfection.
A blueprint for mission-ready AI
The FDA’s story reveals that safe AI in authorities isn’t about bolting fashions onto legacy techniques. It’s about constructing the ruled information basis, safe structure, and operational self-discipline required to make AI helpful in high-stakes environments.
For public sector leaders, which may be an important takeaway of all. When governance is in-built from the beginning, modernization does greater than cut back technical debt. It creates the inspiration to place AI into the arms of scientists, analysts, and decision-makers securely and responsibly in help of the mission.
Be taught extra
Try the session How the FDA Is Scaling Safe AI with Databricks for Authorities introduced at Knowledge + AI Summit, 2026
