AI agent discovery still leaves a governance gap on endpoints

At a glance

What this is: NHI Mgmt Group's analysis, drawing on Lasso Security's blog post: Endpoint-based AI agent discovery remains incomplete unless teams consolidate web, desktop and local AI signals into one governed inventory.

Why it matters: IAM, IGA and security teams need more than visibility because unmanaged AI agents create governance gaps across endpoints, browsers and developer workflows.

Context

AI agent discovery is the process of identifying where generative and agentic tools are actually running, who is using them, and what access they carry. In this article, Lasso Security argues that discovery remains fragmented across browsers, endpoints and desktop applications, which leaves governance teams with incomplete inventory and weak control over shadow AI.

The governance gap is not just about finding tools faster. It is about distinguishing true agentic or generative AI from ordinary software, then attaching usage context, risk and policy treatment so the inventory can support IAM, IGA and security decisions across the estate.

Key questions

Q: What breaks when AI discovery is limited to browser sessions?

A: You miss shadow AI, local copilots, IDE extensions, and many agentic interactions. That creates blind spots in policy enforcement and auditability because the security team sees only a subset of the actual AI footprint, not the full set of places where sensitive data and decisions move.

Q: Why does AI visibility not automatically solve governance risk?

A: Visibility only shows that a tool exists. Governance depends on knowing whether it is generative or agentic, what permissions it uses and how it behaves at runtime. Without that context, teams can see the asset but still fail to assign the right policy, risk score or access treatment.

Q: How do security teams know whether an AI agent is operating safely?

A: Security teams know an AI agent is operating safely when its permissions, invoked tools, and accessed data remain consistent with the approved use case over time. Useful signals include restricted data exposure, unchanged guardrails, and a stable identity path. If any of those drift, the agent should be re-reviewed before it expands further.

Q: How should security teams govern local AI agents that run on developer endpoints?

A: Treat them as NHIs with identity, access, and lifecycle ownership. Inventory each agent, define which user context it inherits, limit its reachable systems, and require approval for commands that modify code, secrets, or network state. If the agent can execute in the same context as the developer, it needs the same level of governance as other privileged machine identities.

Technical breakdown

Why fragmented discovery creates shadow AI blind spots

Fragmented discovery means no single control plane sees browser-based SaaS agents, local desktop assistants, IDE plugins and homegrown AI tools at the same time. When telemetry sits in separate tools, teams can detect isolated events but cannot assemble a complete inventory or assign consistent governance. That breaks the basic assumption that AI use can be governed from a central list of approved applications. In practice, shadow AI thrives in the gaps between endpoint telemetry, web discovery and user context, especially where desktop tooling is installed outside formal procurement.

Practical implication: build discovery coverage that spans browser, endpoint and developer tooling rather than relying on a single visibility source.

Why visibility alone does not equal discovery

Visibility tells you a process or connection exists. Discovery tells you whether the observed tool is actually generative AI, agentic AI, or just adjacent software that should not be treated the same way. That distinction matters because governance decisions depend on context, including permissions, data handling and how the tool behaves at runtime. Without classification logic, organisations can flood their inventory with noise or miss the tools that matter most. Context is what turns raw telemetry into an inventory that can support policy and risk decisions.

Practical implication: classify observed tools by function, privilege and data use before feeding them into governance workflows.

How runtime risk scoring changes AI governance

Runtime insight adds the missing behavioural layer. A discovered AI agent may be low risk when idle but materially different when it is used with broad permissions, sensitive data access or anomalous behaviour. That is why AI governance cannot stop at install-time inventory. It must track how agents are used, which tools they invoke and whether their runtime patterns violate policy. The real control gap is not just discovery, but the inability to connect presence, privilege and behaviour in one place.

Practical implication: combine discovery with runtime monitoring so risk decisions reflect actual usage, not just tool presence.

NHI Mgmt Group analysis

Unified AI discovery is now a governance requirement, not a convenience feature. The article shows that browser agents, desktop assistants and IDE plugins can no longer be treated as separate visibility problems. When the estate is split across channels, governance becomes inconsistent by design. Practitioners should treat unified inventory as the baseline for any AI control programme.

Visibility without context creates false confidence. Seeing an AI tool is not the same as understanding whether it is generative, agentic or business-critical. That is why context about permissions, data handling and runtime use must sit alongside discovery. Otherwise, security teams end up with an inventory that looks complete but cannot drive policy.

Endpoint telemetry is becoming the front line of AI governance. The article's emphasis on local desktop agents and IDE plugins shows where shadow AI is most likely to evade browser-only controls. That shifts the governance boundary toward endpoint-derived signals and away from web-centric assumptions. Practitioners should expect discovery programmes to converge with endpoint security workflows.

Runtime behaviour is the deciding variable in AI risk treatment. A tool's presence matters less than what it can do during execution. If an AI agent has broad permissions, touches sensitive data or behaves anomalously, the governance model must reflect that operational reality. Security teams should stop equating inventory with control.

Discovery gaps expose a larger lifecycle problem. The article points to a broader truth: AI tools are being adopted faster than they are being onboarded into governance, risk and monitoring processes. That means the control issue is not just finding shadow AI, but establishing a lifecycle that keeps discovery, risk scoring and oversight aligned as usage changes. Practitioners should design for continuous reassessment, not one-time inventory.

From our research library:

• 69% of security leaders agree identity management must fundamentally shift to address agentic AI systems, according to the 2026 Infrastructure Identity Survey.
• 53% of security leaders expect AI to run major portions of their infrastructure autonomously within the next three years, according to the 2026 Infrastructure Identity Survey.
• Read next: Shadow AI and AI Agent Discovery Guide

What this signals

AI discovery now starts at the endpoint, not the browser. Local assistants, IDE plugins and desktop agents are where visibility breaks down first, which means governance teams need endpoint-derived telemetry before they can claim a complete inventory. The practical shift is from finding AI tools to continuously reconciling where they actually run.

Shadow AI is often a classification problem before it is a monitoring problem. If teams cannot distinguish generative or agentic tools from ordinary software, their inventory will be noisy and their controls will be misapplied. Discovery programmes need a classification layer that turns raw signals into governable identity and access decisions.

For practitioners

• Consolidate discovery sources Pull browser telemetry, endpoint signals and developer tooling data into one inventory so AI agents are visible across web and desktop use cases.
• Classify AI tools by behavior Separate generative and agentic AI from ordinary software by evaluating permissions, data handling and how the tool behaves at runtime.
• Add runtime risk scoring Score discovered tools using live usage patterns, sensitive-data exposure and anomalous behaviour rather than relying on installation status alone.
• Close shadow AI blind spots Review where local desktop agents and IDE plugins bypass existing visibility controls, then update monitoring coverage for those endpoint paths.

Key takeaways

• Fragmented discovery leaves AI agents outside a single governance view, especially when browser, desktop and developer tools are monitored separately.
• Context matters as much as visibility because governance depends on permissions, data handling and runtime behaviour, not just on tool presence.
• A workable AI control model starts with unified inventory and then adds classification, scoring and monitoring so discovered tools can be governed consistently.

Key terms

• AI Agent Discovery: AI Agent Discovery is the process of finding and cataloging software agents that operate with some independence across systems, data, and workflows. It identifies where agents run, what they can access, which identities they use, and how they connect to tools, APIs, and other services for governance and security.
• Shadow AI: AI agents, copilots, or connected tools operating without full visibility or governance from security teams. Shadow AI becomes an identity problem when those systems authenticate with unmanaged tokens, service accounts, or OAuth apps that can reach production resources.
• Continuous Risk Scoring: Continuous risk scoring assigns and updates an identity risk value as behaviour, entitlements, location, and privilege level change. It gives security teams a prioritisation mechanism for access decisions, but it only works when the score is tied to a clear governance action.
• Endpoint Telemetry: Endpoint telemetry is the process and event data collected from managed devices to understand local activity. For shadow AI, it includes process spawns, file access, and direct outbound calls from the operating system, giving security teams visibility that network tools and SaaS logs often miss.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity security are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an IAM programme, it is worth exploring.