Artificial intelligence is transforming how government agencies operate and fulfill their missions, from streamlining internal functions to enhancing public services. In 2024, federal agencies reported over 1,700 AI use cases, more than doubling the total from the previous year. These projects span everything from cybersecurity and procurement to healthcare and benefits adjudication.
But as the scale of these initiatives grows, so does the importance of trusted, well-governed data. AI doesn’t succeed on algorithms alone, it runs on data. The effectiveness, safety, and compliance of AI systems depend on how that data is curated and governed.
The Surge for AI Within Government
Government AI adoption is accelerating at an unprecedented pace. According to the most recent federal AI use case inventory:
- 37 agencies are actively developing or deploying AI.
- Mission-enabling functions (finance, HR, IT, cybersecurity) account for 46% of reported use cases.
- Health and medical applications make up 13%, with growing use in fraud prevention, public health surveillance, and case decision support.
Agencies cite benefits like enhanced anomaly detection, streamlined processes, and improved decision-making. For example, the CDC is using AI to accelerate investigations of multistate foodborne outbreaks, while the Social Security Administration is using it to support faster, more fair Disability Program determinations.
Every one of these use cases relies on curated, compliant data with proper access controls for sensitive information.
Mission-Driven AI Starts with Data
Data is the core ingredient of any AI system, yet many agencies still face challenges with fragmented data architectures, legacy silos, and limited metadata management. Even as agencies deploy AI at scale, the question remains: Is the data AI is learning from accurate, complete, and governed?
Agencies have made strides in building enterprise data and analytics platforms, and over 35% of AI use cases now leverage existing production-level code or data. Many are developing documentation for their AI use cases outlining the appropriateness of data for analysis and decision-making.
To avoid biased or non-compliant AI systems, agencies must standardize how they manage and control data across its lifecycle, from ingestion and classification to access and retention.
In-House Models Require In-House Data Discipline
Nearly half of federal AI systems are developed in-house, reflecting the growing AI capacity across agencies. Custom models are accelerating adoption and helping tailor solutions to agency-specific needs, but they also increase the need for data governance.
Unlike commercial software, in-house AI systems require internal teams to curate training data and ensure ethical use of outputs. Without consistent, organization-wide standards for data governance, access control, and defensible ROT deletion, even the most promising AI models risk producing untrustworthy results.
The data isn’t just fuel for a model, it’s the mission-aligned foundation of the model’s integrity.
Unstructured Data: The Next Frontier for AI in Government
While structured data systems have traditionally powered digital initiatives, the real opportunity for government AI lies in the massive volumes of unstructured data — emails, documents, case files, transcripts, reports — that remain largely untapped.
Emerging technologies like Generative AI, semantic search, and advanced retrieval-augmented generation (RAG) are unlocking new ways to analyze and extract insights from unstructured content. However, without proper data governance, this content introduces serious risk, from regulatory exposure to privacy violations.
To safely harness unstructured data, agencies must:
- Index and classify documents with content and metadata at scale.
- Establish retention policies and automate defensible deletion.
- Apply role-based access controls to limit who (or what) can interact with sensitive content.
- Ensure explainability and auditability of any AI that draws from this data.
Governance Guardrails at The Data Layer
As the Office of Management and Budget (OMB) emphasized in Memorandum M-24-10, federal AI systems must be designed to protect the rights and safety of the public. Of the 1,700+ reported use cases last year, 13% are considered rights-impacting, triggering strict requirements around risk assessments, independent evaluations, and the ability to opt out in favor of a human decision-maker.
There’s another crucial guardrail that often goes overlooked: secure access control. Government agencies handle some of the most sensitive information in the world, including health records, classified intelligence, and veterans’ benefit data. If left ungoverned or improperly governed, this information could be exposed to unauthorized users or leaked into external AI models.
That’s why any agency data accessed by an AI system must be subject to rigorous access and usage controls, ensuring that only authorized personnel (or AI agents) can interact with it, and only for approved purposes.
Data governance solutions enable this by:
- Enforcing zero-trust access frameworks.
- Providing granular permissions down to the document or field level.
- Creating audit trails for every access and inference.
- Ensuring AI models are not trained on restricted content without explicit approval.
These controls are essential for building trustworthy, mission-driven, and legally defensible AI.
The Data Governance Imperative
AI offers immense potential for mission-driven government deployments, from faster services to smarter policymaking. As agencies develop their own models and tap into the intelligence contained in unstructured data, it has become clear that AI success depends on enterprise-grade data governance.
Ready to leverage your agency’s unstructured data to power mission-driven, trustworthy AI? Contact ZL Tech today.