AI Query Translation Eliminates Government Data Exposure Risk

A new approach to artificial intelligence in government is emerging that fundamentally sidesteps the privacy concerns that have stalled many agencies. The concept is simple: use AI as a translator rather than a processor. Instead of feeding sensitive data into a model, the AI converts human language into database queries that run on existing secure systems. The model never sees the actual information.

This architecture was demonstrated at the National Association of State CIOs Midyear Conference in Philadelphia by Scott Drzyzga, executive director of Pennsylvania's State Geospatial Coordinating Board. He showcased an implementation from Crawford County in northwestern Pennsylvania that added an AI chat feature to its parcel viewer application. The system allows citizens to ask complex questions about property data without exposing that data to the AI model itself.

During the demonstration, Drzyzga selected a specific property and prompted: "find properties within a half-mile that sold for more than $1 million in the last three years." The AI interpreted this compound query into SQL code, which was then executed against the county's existing database. The results returned as a table with a mini-map visualization. The AI agent never touched the owner information, never saw the addresses, never accessed the property values, and never stored any data.

That distinction matters enormously for government agencies navigating public records laws and privacy regulations. As govtech.com reported, many agencies have been explicitly purchasing enterprise versions of AI platforms that don't feed inputs back into larger public models for training. California agencies have gone further, with staff guidelines stating that nothing subject to a public records request should be run through generative AI at all. This architectural approach essentially eliminates those concerns.

The technical implementation is elegant in its simplicity. The AI operates at the user interface layer, enabling anyone to work with database queries regardless of technical ability. The core functionality of the parcel viewer application remains unchanged. What changes is accessibility. Citizens who would never realistically construct a complex SQL query can now ask questions in plain English. The friction of learning database syntax disappears (which is probably why most people just click through menus anyway).

This pattern represents a broader shift in how government agencies are thinking about AI deployment. Rather than treating AI as a replacement for existing systems, agencies are positioning it as an interface layer that democratizes access to data without compromising security. The physical experience for end users shifts from navigating dropdown menus and clicking through filters to simply typing a question and waiting for results to populate on screen.

Independent reporting from Government Technology documents similar approaches across multiple jurisdictions. Washington, D.C. uses real-time dashboards for blood supply management in field transfusions. Ohio's RecoveryOhio Overdose Early Warning Dashboard predicts ZIP codes with increased overdose risk up to 30 days in advance. These systems rely on strong data governance and infrastructure modernization to support AI adoption while maintaining security boundaries.

The Crawford County example highlights a critical architectural decision: where does the AI sit in the data pipeline? Traditional implementations feed data into models for analysis, creating exposure risks. The query-translation model keeps sensitive data in existing secure systems and only exposes the query structure to the AI. This is a fundamental difference in threat surface.

State-level AI legislation is increasingly addressing these concerns. Texas Senate Bill 1964 created a regulatory structure for transparent government AI use. The Texas Responsible Artificial Intelligence Governance Act requires explicit consent for commercial use of biometric data. California Assembly Bill 502 prohibits malicious AI-generated media in election communications. These laws create the policy framework that technical implementations like Crawford County's must operate within.

However, gaps remain in federal guidance. A GAO report from January 2026 found that the Office of Management and Budget's government-wide AI guidance doesn't fully address major privacy-related risks and challenges. The report identified ten expert-identified privacy challenges, with OMB guidance addressing only two of them. Without additional direction, risks increase that agencies' use of AI would disclose sensitive data or compromise privacy in other ways.

The White House is also studying an executive order requiring pre-deployment review of frontier AI models. Kevin Hassett, director of the National Economic Council, compared the approach to how the Food and Drug Administration evaluates drugs for safety. This would likely increase workload at NIST's Center for AI Standards and Innovation, which has already conducted 40 evaluations including on unreleased models. The center received agreements with Google DeepMind, Microsoft, and xAI to conduct pre-deployment evaluations.

Resource constraints complicate these efforts. The America First Policy Institute called CAISI "chronically underfunded" with approximately 30 total staff and $30 million since establishment in 2024. The Federation of American Scientists advocated for annual operating budgets up to $155 million plus $155-275 million in setup costs for high-security compute facilities. Whether the center can adequately carry out its mission with current resources remains uncertain.

State governments are moving forward regardless. A Council of State Governments report indicates AI has been a top legislative priority for states in 2025, with 252 AI-related measures proposed. Only 12 states have not yet enacted AI legislation. Public-private partnerships are accelerating implementation, including Amazon's investment in AI infrastructure in Pennsylvania and California's partnership with NVIDIA for AI workforce training.

The Crawford County implementation demonstrates that technical solutions exist for the privacy concerns that have paralyzed many agencies. The question becomes whether agencies have the technical capacity to implement these architectures and the policy frameworks to govern them properly. Data governance practices vary widely across jurisdictions, with most states lacking data quality programs and only about half appointing chief data officers.

Whether this architectural approach scales beyond parcel viewers to more sensitive government data remains to be seen. The technical feasibility is proven. The policy framework is still catching up. And the real test will be whether agencies can maintain these security boundaries as they expand AI capabilities across more mission-critical systems.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn