As AI scales in government, visibility, zero trust and data protection are critical

State and local governments are accelerating their use of artificial intelligence to drive innovation, improve efficiency and enhance citizen experience. Across the country, state and local governments are already putting AI to work. 

The North Carolina Department of State Treasurer has integrated AI into daily workflows to boost productivity and better steward taxpayer dollars. In New York, more than 100,000 state employees are being trained to use AI responsibly to improve how they serve residents. Cities and towns are turning to AI to tackle permitting backlogs and help address the housing crisis.

But as adoption accelerates, so do the risks.

AI introduces new challenges that state and local governments are working to understand — from inaccurate or unverifiable outputs to the growing use of autonomous systems that can act on sensitive data. Incidents like hallucinated legal filings and emerging concerns around Anthropic’s Mythos underscore a broader reality: without the right oversight, AI can introduce errors and expose sensitive data at scale.

To safely adopt AI at scale, agencies need to balance innovation with security — grounded in visibility, zero trust and continuous data protection.

Visibility Gaps in AI Inventories Are Creating Risk

While AI adoption is accelerating, visibility is not necessarily keeping pace. Recent research reveals AI usage in critical sectors has grown by more than 200%, yet many organizations still lack a basic inventory of the AI tools and models in use. At the same time, AI systems can be compromised in as little as 16 minutes.

Without a clear inventory of AI tools, agencies can’t answer the following questions: What data is being used? Where is it stored? How is it being accessed? And what AI tools have access to what? As AI usage expands across models, agents, development environments and embedded software-as-a-service features, limited visibility leaves security teams unable to fully assess exposure or enforce policy.

This also fuels “shadow AI,” where teams adopt tools faster than security can standardize access, logging and data protections — creating new pathways for sensitive data exposure, policy violations and operational risk.

Establishing a comprehensive view of the AI, across models, agents, applications and data flows, is the first step toward safe AI usage. It allows agencies to understand how AI systems interact with sensitive data and prioritize risk based on real-world usage patterns.

Just as important, visibility enables accountability. Agencies can define what data is being used for AI, why it is retained and where guardrails are needed.

Zero Trust: Verifying How AI Interacts with Systems and Data

While visibility is an important step, it is not the end goal. Even with full visibility, traditional security models fall short in an AI-driven environment.

Traditional, perimeter-based security models rely on implicit trust, assuming that once a user is inside the network, they can operate freely. Zero trust, on the other hand, removes implicit trust altogether. Built on the principle of “never trust, always verify,” zero trust continuously evaluates every interaction – whether human or machine – based on identity, context and policy. It enforces least-privileged access, ensuring AI systems can only access the data and services it is explicitly authorized to use.

AI doesn’t behave like traditional software. It learns, adapts and interacts with data in dynamic ways. As systems become more autonomous, risk expands beyond external threats. AI can introduce unintended consequences from within — accessing data, interacting with systems and acting in ways no one explicitly designed.

For state and local governments, this is essential. Whether it’s a benefits system determining eligibility or a municipal platform processing permits, zero trust ensures that access is tightly controlled and continuously validated.

In an environment where AI can act independently, zero trust becomes the mechanism that governs those actions, reducing the risk of unintended access, lateral movement and large-scale data exposure.

Continuous Data Protection

Continuous data protection also ensures what matters most stays secure. AI systems are powered by data, and often, that data includes highly sensitive information about citizens. Without the right safeguards, that data can be exposed through prompts, outputs, or integrations with external tools. For state and local agencies, data exposure can lead to legal liability, regulatory violations and erosion of public trust.

That’s why continuous data protection, paired with red teaming and visibility into data traffic flows, is vital. Agencies need to be able to inspect and control how data is used in real time — preventing sensitive information from being shared with unauthorized AI services, applying policies dynamically and monitoring for misuse or anomalies.

Red teaming plays a key role by proactively testing how AI systems behave under real-world conditions. By simulating misuse, adversarial inputs and edge cases, agencies can identify vulnerabilities and unintended behaviors before they surface. Combined with runtime guardrails and prompt filtering, this approach reduces risk while allowing agencies to safely scale AI use.

Maintaining visibility into where data is moving is essential. Governments operate systems that determine benefits eligibility or citizen services, oversight is critical to keep critical life and safety services operating.

As AI continues to evolve, these controls must evolve with it. Security needs to remain adaptive, combining continuous data protection and active testing to ensure sensitive data stays protected as use cases expand.

A Secure Path Forward for State and Local AI

AI is already reshaping how state and local governments serve their communities. As the technology continues to advance and adoption scales, agencies must have visibility into their AI inventory, implement zero trust and continuously protect their data.

These steps remove blind spots, reduce risk and maintain control as AI becomes more embedded in daily operations. In the public sector, that control is critical. It allows agencies to scale AI securely, protect sensitive data and strengthen the resilience of the services that citizens depend on every day.

Adam Ford is the chief technology officer for state and local government and education at Zscaler. Prior to Zscaler, Adam served as Illinois chief information security officer for five years. In this role, he had responsibility for cybersecurity for agencies, boards and commissions under the governor.