Autonomous Systems Introduce New Operational Risk
Agentic AI can initiate actions, access data and interact across environments without direct human input. In government environments, where systems are interconnected and data is widely shared, this type of behavior creates operational risk.
AI systems supporting services such as benefits or permitting may interact with multiple systems containing sensitive information. Without clear constraints, unintended actions can affect both security and service delivery, directly impacting citizen services.
One way agencies can better protect sensitive data and maintain compliance is by transitioning to a zero-trust architecture.
READ MORE: IT modernization supports transformational government.
Legacy Architectures Create Risk and Complexity
Most state and local governments still rely on perimeter-based security models, such as VPNs. These models grant broad access once users are inside the network, and from there, users are left unchecked.
Recent research highlights how reliance on legacy security models is increasing exposure to attacks. For state and local governments, this puts sensitive data and critical citizen services at risk. As attackers use AI to move at machine speed, many defenders still rely on legacy VPN architectures that limit visibility and slow response. According to the report, 70% of the organizations surveyed report limited or no visibility into AI-driven threats moving over VPN traffic.
These findings underscore a broader issue: Agencies must have continuous visibility into their environments. Perimeter-based models assume trust once inside the network, allowing users — human or machine — to move laterally without sufficient controls.
Zero trust removes implicit trust by ensuring users and systems access only what they need, when they need it. By enforcing least-privileged access and continuously verifying interactions, agencies can limit lateral movement and contain risk when systems or internal users behave unexpectedly.
For state and local governments, this shift may seem daunting given limited resources, budgets and time. However, zero trust does not require a complete overhaul. Many agencies can begin by replacing VPN-based access for remote users or high-value applications, then expand controls incrementally based on risk and mission-critical priorities. This phased approach aligns with limited resources and ongoing modernization efforts.
Building Visibility Into AI Systems and Data
While implementing zero trust is essential, agencies also must understand where AI is being used and how it interacts with data. AI capabilities are increasingly embedded across applications and workflows, often without centralized oversight.
Without visibility, agencies cannot identify where sensitive data is being shared or how systems are interacting. Establishing visibility across applications, prompts and data flows allows agencies to detect shadow AI, enforce policy and protect critical information while also improving governance and accountability across departments.
LEARN MORE: State and local agencies must beware shadow AI.
Continuously Validating AI Behavior
Because AI systems operate continuously, security must operate continuously as well.
Testing systems before deployment, monitoring behavior in real time and enforcing controls during interactions help identify vulnerabilities and reduce risk. Approaches such as red teaming and runtime monitoring ensure systems behave as intended, even as conditions change.
As these systems scale, manual oversight alone will not be enough. Security teams will need to use AI to keep pace — applying automation to detect anomalies, uncover risky behavior and respond in real time. This “fight AI with AI” approach reflects how security must evolve alongside the technology it is designed to govern.
Strengthening Security While Modernizing Government IT
State and local governments are under pressure to modernize services, improve the digital experience and maintain critical operations, all while managing limited resources.
AI can support these goals, but only if it is deployed with the right safeguards. By reducing exposure with zero trust, building visibility into AI use and continuously validating system behavior, agencies can strengthen their security, support modernization and protect the services citizens rely on every day.
