# Why federal agencies need to rethink trusted access in the age of AI > Source: > Published: 2026-10-02 15:42:51+00:00 # Why federal agencies need to rethink trusted access in the age of AI As federal agencies increasingly deploy more AI-powered agents alongside traditional AI tools, the insider threat landscape is becoming harder to manage. Employees, contractors, applications, and systems all require legitimate access to sensitive data and resources, but AI is creating new ways for that authorized access to be misused, compromised, or extended beyond its intended purpose. An employee may move sensitive information into an unapproved AI tool, an AI agent may access or share data beyond its intended role, and an attacker using stolen credentials may still appear to be a legitimate user. These scenarios are different, but they create the same underlying risk: trusted access being used in ways agencies did not intend. The Cybersecurity and Infrastructure Security Agency 2026 [Insider Threat Mitigation Guide](https://www.cisa.gov/sites/default/files/2026-09/cisa-insider-threat-mitigation-guide-2026.pdf) reflects that changing landscape, emphasizing the need to prevent the exploitation of authorized access and help organizations identify and respond to insider risk. For agencies, that means knowing where trust extends, validating how it is used, and limiting its reach when something goes wrong. ## **Understand where trust exists** Agencies cannot manage insider risk without understanding how access moves across their environment. Security teams need network visibility into which users, applications, and AI agents can connect to sensitive resources, how those resources communicate, and where privileges extend farther than necessary. A [recent bulletin](https://www.dcsa.mil/Portals/128/Documents/CI/DITMAC/BTAC%20BULLETIN%20JUNE_2026.pdf) from the Defense Counterintelligence and Security Agency’s Behavioral Threat Analysis Center found that 44% of organizations surveyed had minimal to no visibility into AI agent activity. DCSA also warned that shadow AI can create new paths for data exposure and operational security failures. As AI tools become more widely available, agencies should continuously map communication and access patterns rather than rely on a one-time inventory. That means identifying high-value assets, determining which connections are mission-essential, and flagging new or unnecessary pathways that could indicate misuse or exposure. ## **Monitor behavior, not just authorization** Visibility establishes the baseline. Agencies then need to verify whether users, applications, and other identities are behaving as expected. A valid login does not guarantee that the activity that follows is safe. An attacker using stolen credentials may initially look like the employee whose account was compromised. Warning signs should surface when that identity connects to unfamiliar resources, operates outside its normal workflow, or attempts to access sensitive areas. The same principle applies across the insider threat spectrum. A malicious insider may look for information outside normal responsibilities. A negligent user may move or share data in ways that create exposure. A third party with legitimate access may begin interacting with systems unexpectedly. Agencies should establish normal communication and usage patterns and investigate unusual activity. While authorization answers whether an identity can access a resource, continuous verification helps determine whether that activity makes sense in context. ## **Design for the failure of trust** Even strong visibility and monitoring will not catch every suspicious action before damage occurs. That is why agencies should also design for the failure of trust. Once legitimate access is compromised or misused, the key question becomes how far the impact can spread. CISA’s expanded focus on access control is relevant here. Segmentation can place boundaries between applications, workloads, and critical systems. Rather than allowing users, devices, or applications to communicate broadly, agencies must restrict connections to those required for a role or mission. In practice, that means creating policies that allow required traffic while blocking unnecessary connections. If credentials are stolen, segmentation can restrict lateral movement. If an unauthorized application appears, agencies can limit what it can communicate with. If suspicious activity is detected, containment can give security teams more time to investigate without allowing the incident to spread. Those controls become even more important as agencies introduce AI agents that may act across multiple environments on a user’s behalf. Limiting an agent to the resources it actually needs can reduce the blast radius if that access is misused or compromised. ## **Rethinking the insider threat model** Training, reporting, and behavioral indicators will remain essential parts of insider threat prevention. Employees still need to recognize suspicious behavior, handle sensitive information appropriately, and use emerging technologies responsibly. But those measures should be paired with technical controls that reduce the risk created by trusted access itself. As identities, applications, and AI systems become more interconnected, agencies may not be able to prevent every misuse or compromise. They can, however, design their environments so that one stolen credential, mistake, or unexpected action does not become a mission disruption. The rise of AI is forcing agencies to rethink a longstanding security assumption: that trusted access is inherently safe. As AI agents, applications, and users become more interconnected, trust can no longer be treated as a static state. Agencies that pair visibility with continuous verification and containment will be better positioned to manage insider risk, whether it originates from human error, malicious intent, or the unintended consequences of AI-driven activity. *Gary Barlet is the public sector CTO for Illumio. He previously served as CIO for the U.S. Postal Service’s Office of Inspector General.*