TL;DR: AI governance breaks when discovery cannot see the tools people actually use, according to Lansweeper, because approved-tool lists and scoped scans miss browser extensions, local models, and unmanaged connections. Evidence-based inventory is now the control that determines whether policy, audit, and automation can be trusted.
At a glance
What this is: NHI Mgmt Group's analysis, drawing on Lansweeper's blog post: AI governance is failing where discovery only covers approved assets and misses the AI tools people actually use.
Why it matters: Identity, access, and security teams cannot govern AI use, evidence retention, or policy enforcement if the inventory is built from assumptions instead of observed runtime activity.
By the numbers:
- 69% of organisations suspect or have evidence that employees are using prohibited public GenAI tools.
- Lansweeper says its AI usage tracking captures external AI services over a 14-day window tied to individual assets.
Context
AI governance breaks first at the inventory layer. Policies can name approved tools, but they cannot govern what is not discovered, and traditional asset discovery often misses browser extensions, local models, and unmanaged AI connections. That creates a control gap between policy intent and actual runtime use.
The problem becomes sharper when governance depends on evidence. Risk classification, documentation, and audit response all assume teams can enumerate the systems in play, yet many environments still rely on sanctioned lists rather than observed usage. Where AI is spreading through endpoints and browsers, inventory is now a governance prerequisite, not an administrative output.
For identity and access practitioners, the intersection is direct: AI tools can move data, consume entitlements, and create shadow access paths outside the normal approval chain. AI governance guide concepts matter here because discovery, ownership, and evidence are now part of access governance, not just compliance bookkeeping.
Key questions
Q: How can organisations detect and contain shadow AI agent access?
A: Organisations should correlate network activity, system logs, and endpoint evidence to find access that never appeared in identity telemetry. Once detected, they should block the unauthorized path, revoke any active session or account, and record the incident in a workflow that preserves evidence for audit and response.
Q: Why do approved AI policies fail in practice?
A: Approved AI policies fail when they are written as static rules but not enforced against real model use, data flow and tool access. Without runtime visibility, policy cannot detect shadow AI, unauthorised prompts, unsafe data sharing or agents acting outside approved scope. Effective controls combine inventory, gateway logging, prompt and output monitoring, access restrictions and exception handling.
Q: What is the difference between AI inventory and AI usage tracking?
A: AI inventory is the catalogue of AI systems, models, agents and related services in use, with owners, purposes and risk context. AI usage tracking records when, how and by whom those systems are actually used, including prompts, API calls, connected tools and data access. Inventory tells you what exists. Usage tracking shows behaviour and policy compliance.
Q: What should teams do if they discover shadow AI in the business?
A: Teams should first identify who owns the tool, what data it touches, and which identities it uses. Then they should either bring it under policy and lifecycle control or remove access to enterprise data until governance is in place. Discovery without containment simply confirms the scale of the gap.
Technical breakdown
Why approved-tool lists miss actual AI usage
Approved-tool lists describe what an organisation intended to allow, not what employees actually used. Browser extensions, unmanaged laptops, contractor devices, and local model servers can all bypass a discovery model that starts from predefined scopes such as subnets or asset groups. The technical failure is not visibility in general, but scope blindness: if a scanner only interrogates known inventory, it will never identify unknown AI services already exchanging data with endpoints. See the AI Governance Platform Buyer's Guide.
Practical implication: Treat AI discovery as a runtime visibility problem and measure whether your tools can detect use outside pre-approved inventory scopes.
How AI usage telemetry turns activity into evidence
Usage telemetry changes governance from a static allow list to a time-bound record of observed connections. When a control captures which external AI services a device reached, how often it connected, and what assets generated the traffic, it creates evidence that can support policy enforcement and audit review. The key technical shift is from configuration truth to behavioural truth, which matters because AI use often emerges through the browser rather than through centrally deployed software. See the ISO/IEC 42005 AI System Impact Assessment Guide.
Practical implication: Collect endpoint and network telemetry that records observed AI service connections and retains enough context to support audit questions.
Why network-based discovery changes the governance baseline
Network-based discovery does not wait for assets to be added to a management scope. It infers presence from traffic, so an active AI tool on an unsanctioned device becomes visible because it is communicating, not because someone remembered to list it. That matters for governance because inventory gaps become policy gaps when automation, review workflows, and compliance reporting all consume the same incomplete source of truth. The real control issue is not just unknown software, but unknown activity that keeps moving data. See the AI Governance Guide.
Practical implication: Use network-observed discovery to close blind spots created by pre-scoped scans and to surface active AI use on unmanaged devices.
NHI Mgmt Group analysis
AI governance debt is now an inventory problem before it is a policy problem. Organisations can write acceptable-use rules and still remain blind to the tools employees actually run. When the inventory is incomplete, the control framework is managing intention rather than reality, and that gap widens as AI becomes embedded in browsers, endpoints, and developer workflows.
Shadow AI is the governance analogue of shadow IT, but it moves faster and leaves less administrative trace. An AI tool can be installed, connected, and used before any central team sees it, which means the governance burden shifts from approval to detection. The practitioner conclusion is straightforward: if discovery cannot prove presence, policy cannot prove compliance.
Runtime evidence is becoming the audit currency for AI governance. The organisations that will withstand board and regulator questions are the ones that can show how AI use was observed, not just how it was permitted. That pushes AI governance toward monitored activity, retention of usage evidence, and explicit ownership of discovered tools, rather than reliance on static allow lists.
Identity governance now extends to the AI services that consume enterprise data. Once a browser extension or local model can move information into an external service, the question is no longer only whether the user had permission. The harder question is whether the organisation can see, classify, and constrain the service itself across the full data path.
From our research library:
- Only 44% of organisations have implemented any policies to manage their AI agents, despite 92% agreeing that governing AI agents is critical to enterprise security, according to the 2026 Infrastructure Identity Survey.
- Read next: AI Governance Platform Buyer's Guide
What this signals
Shadow AI governance will move from policy authorship to runtime verification. The teams that still depend on approved-tool registers will keep finding that their control plane lags actual usage, especially where browsers and local models create unsanctioned data paths. When 52% of respondents see AI security decision-making power shifting toward platform and infrastructure teams rather than the executive suite, governance naturally follows the operational telemetry that those teams can actually measure.
Evidence, not intent, will become the new standard for AI control decisions. Boards and auditors will increasingly ask for observed connections, retention windows, and ownership of discovered services rather than policy statements alone. That pressure will push programmes toward continuous discovery, because static approvals cannot prove what is running on unmanaged endpoints or who is exchanging data with external AI services.
For practitioners
- Define AI inventory as a runtime control Require your AI inventory to reflect observed usage, not just approved tools, and include browser activity, endpoint software, and unmanaged connections in scope.
- Separate approval from detection workflows Maintain one register for sanctioned tools and a second operational view for discovered AI services so gaps are visible instead of hidden inside a single approved list.
- Retain evidence for audit and incident response Keep time-bound records of which devices connected to which AI services, how often they connected, and what data paths were involved so auditors can verify governance claims.
- Treat unmanaged AI connections as governance exceptions Route unsanctioned AI use into the same exception handling process you use for risky software or unauthorized cloud services, with ownership and closure dates.
Key takeaways
- AI governance fails when discovery is limited to approved assets and cannot see the tools people actually use.
- Static policies are not enough when unmanaged browser extensions, local models, and external AI connections can move data outside normal oversight.
- Runtime discovery and evidence retention are now core governance controls for proving what AI is present, how it is used, and who owns it.
Related NHI Mgmt Group resources: AI Governance Guide · ISO/IEC 42005 AI System Impact Assessment Guide · AI Governance Platform Buyer's Guide · Singapore Model AI Governance Framework for Agentic AI Guide
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
MITRE ATT&CK addresses the attack surface, NIST AI RMF and NIST CSF 2.0 set the technical controls, and ISO/IEC 42001:2023 defines the regulatory obligations.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST AI RMF | GOVERN — AI Governance and Accountability | AI governance debt here is fundamentally a governance and accountability problem around undiscovered AI use. |
| Recommendation — Establish governance ownership for discovered AI services and require evidence-based approval records. | ||
| ISO/IEC 42001:2023 | AI management system — AI Management System | The post is about systematic control of AI use, inventory, and evidence inside an AI management system. |
| Recommendation — Define an AI management system that covers inventory, ownership, evidence retention, and review. | ||
| NIST CSF 2.0 | ID.AM-01 — Physical Devices and Systems Inventoried | Discovery gaps mean AI-using devices and services are not fully inventoried before governance begins. |
| PR.AA-05 — Access Permissions, Entitlements and Authorizations | Unmanaged AI services can create unauthorised data access paths that should be governed as entitlement issues. | |
| Recommendation — Maintain a current inventory of devices and systems that generate or consume AI traffic. Review AI-connected services for excess permissions and remove access that is not explicitly authorised. | ||
| MITRE ATT&CK | TA0007;TA0010 — Discovery; Exfiltration | The control gap is that hidden AI tools evade discovery while still moving data out of the environment. |
| Recommendation — Map unmanaged AI traffic to discovery and exfiltration tactics, then hunt for active unapproved services. | ||
Key terms
- 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.
- AI Usage Tracking: AI Usage Tracking is the recording and review of how AI systems are used in practice. It captures prompts, responses, API calls, connected tools, users or agents, and data access so organisations can see actual behaviour, enforce policy and investigate misuse, drift or unauthorised use.
- AI Inventory: An AI inventory is a governed record of all AI-related assets, enriched with owner, purpose, access, and risk context. It turns discovery into something security, compliance, and IAM teams can use to make approval, review, and revocation decisions.
- Network Discovery: Network discovery is the process of identifying devices, services, and connections on a network to understand what is present and how it communicates. In security operations, it combines active probes and passive observation to map assets, reveal exposed services, and support IAM, NHI governance, attack surface management, and incident response.
Deepen your knowledge
NHI Mgmt Group covers identity security, NHI governance, and agentic AI through independent research, practitioner guides, and the NHI Foundation Level course. Explore nhimg.org for resources that connect identity governance to the broader security disciplines your programme depends on.
Published by the NHIMG editorial team on November 9, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org