TL;DR: Many organisations have already adopted AI, but most leaders still cannot map where risk sits across model use, shadow AI, autonomous agents, and data exposure, according to JetStream Security research. The finding is that governance fails when identity, data, and approval paths are not inventoried before use spreads.
At a glance
What this is: NHI Mgmt Group's analysis, drawing on JetStream Security's blog post: AI risk is framed as an inventory and governance problem, with many leaders unable to pinpoint where model use, autonomous agents, and data exposure are actually happening.
Why it matters: It matters because IAM, NHI, and AI governance teams need a defensible view of who or what is acting under which identity before they can set policy, prove control operation, or contain shadow AI.
Context
AI governance breaks down first as a visibility problem, not a tooling problem. Once staff adopt models and agents faster than policy can track them, organisations lose sight of which systems hold data, which identities act on behalf of users, and which approvals were never captured.
The governance gap is broader than one workflow or one department. It spans sanctioned and unsanctioned AI use, delegated access, data handling, and evidence collection, which is why a programme needs a clear inventory before it can claim control over agentic AI identity maturity.
The same problem appears when AI use moves from experimentation to production without an owner for each access path. That is the point at which risk stops being theoretical and starts becoming a governance gap that no blended score can hide.
Key questions
Q: What should teams do first to govern AI use across the organization?
A: The first step is to establish clear acceptable use rules and align them with access governance. Teams should define which AI tools are approved, who can use them, what data they may process, and how requests are reviewed. That creates a baseline for control, makes enforcement possible, and gives security teams a practical starting point for broader AI governance.
Q: Why do shadow AI tokens create more risk than visible automation?
A: Because visibility into output does not equal control over access. An AI workflow can behave normally while reaching cloud, SaaS, or internal systems through a secret that was never intended to be permanent or broadly scoped. The risk is the hidden entitlement, not the task automation itself.
Q: What should organisations rethink when AI agents can act without human approval?
A: Organisations should rethink review cycles, revocation timing, and accountability assumptions. If an agent can complete a task before a human review occurs, then access reviews no longer capture the full risk. Governance has to move to runtime policy, per-agent identity, and machine-speed lifecycle controls.
Q: How do organisations know if AI governance is actually working?
A: They should be able to reconstruct a live interaction from identity context, policy outcome, accessed resources, and enforcement evidence. If the organisation can only show a policy document or a generic alert, governance is incomplete. Working AI governance leaves behind reviewable artefacts that compliance, legal, and security teams can use without guessing what happened.
Technical breakdown
Why AI governance fails when adoption outruns inventory
AI programmes usually fail at the point where use becomes normalised faster than it is documented. The core issue is not that teams lack policies, but that the organisation cannot reliably enumerate models, tools, datasets, agents, and the human owners behind them. Without that inventory, governance becomes reactive: exceptions are discovered after data moves, prompts are stored, or an agent reaches a system it should not have touched. A maturity model only becomes useful when it measures whether the organisation can see the estate before it tries to govern it. See the Identity Security Maturity Model.
Practical implication: Treat AI inventory as the prerequisite control, not the last reporting step.
How shadow AI creates governance blind spots
Shadow AI appears when teams use public or unmanaged tools outside approved review paths. The risk is not limited to policy breach; it includes untracked data disclosure, loss of auditability, and inconsistent retention of prompts or outputs. Because usage often begins in ordinary business tasks, the governance failure is structural: the organisation never gets a complete map of where AI touches client data, legal material, or internal strategy. That makes later enforcement difficult because the programme is trying to govern activity it never formally discovered.
Practical implication: Build discovery and review processes that surface unmanaged AI use before it becomes embedded in business workflows.
Why autonomous agents change the identity model
An autonomous agent is not just another software tool with an API key. When it can select actions, choose tools, and execute without a human approval gate, it behaves as an actor that can expand or chain access during runtime. That breaks the assumption that privilege is fixed at provisioning time. In this context, the relevant governance question is not only who created the agent, but what identities it can assume, what data it can reach, and how its action trail is proven after the fact. This is where AI governance and identity governance converge. See the Agentic AI Identity Maturity Model.
Practical implication: Classify agent identities separately from human users and review their runtime authority as a distinct control domain.
Breaches seen in the wild
- iOS apps leaking hard-coded secrets: Cybernews found 71% of 156,080 iOS apps leak hard-coded secrets, with open cloud storage and Firebase databases exposing user data.
Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.
NHI Mgmt Group analysis
AI governance fails when organisations treat adoption as the starting point and inventory as an afterthought. The programme cannot govern what it cannot name, and this article shows that AI risk quickly spreads across legal, technical, and operational use cases. The practical conclusion is that inventory, ownership, and approval lineage have to be established before the programme can mature.
Shadow AI is a governance defect, not just a policy violation. When employees paste sensitive data into unmanaged tools, the problem is that the organisation has lost control over data flow, evidence, and retention. That means security and legal teams need a shared view of sanctioned use, because policy alone does not reconstruct missing oversight.
Agent identity is now part of AI governance whether the programme acknowledges it or not. Once autonomous agents act under identities no one inventoried, the control problem shifts from user training to runtime authority. That is why identity governance for AI has to cover who owns the agent, what it can do, and what evidence proves it did not exceed scope.
Identity blind spot: The most important failure mode here is not model risk in isolation, but identity risk without inventory. A programme that cannot tie an AI action to a named owner, a known access path, and a reviewable control boundary is already behind. Practitioners should treat that gap as a programme design flaw, not a reporting weakness.
Read next: Identity Security Maturity Model
What this signals
Governance programmes that start with policy rather than discovery will continue to miss the same failure mode: the organisation cannot govern what it has not inventoried. The next step is to turn AI use into a known estate with ownership, approval lineage, and visible access paths, not a collection of disconnected experiments.
Runtime governance gap: autonomous AI changes the control surface from user behaviour to actor behaviour. Once an agent can select actions and act under delegated access, existing review cycles no longer describe what actually happened, so identity governance has to move closer to issuance and execution time.
For practitioners
- Map every AI use case to a named owner Document who approved the tool, which team uses it, what data it can touch, and which identity it runs under before expansion continues.
- Separate sanctioned AI from shadow AI Create a discovery process for public tools, unmanaged browser use, and embedded assistants so unsanctioned activity is visible early.
- Review agent permissions as runtime authority Treat autonomous agents as separate actors with explicit access boundaries, not as extensions of the human user who configured them.
- Capture evidence for data handling and approvals Keep records of prompts, outputs, approval lineage, and data exposure paths so legal, audit, and security teams can reconcile the same event.
Key takeaways
- AI governance breaks down when adoption spreads faster than inventory, leaving organisations unable to name where models, agents, and data exposure actually sit.
- The article’s core signal is that shadow AI and autonomous agents create an identity and evidence problem, not just a policy problem.
- Practitioners need named ownership, discovery, and runtime boundaries before they can claim that AI governance is operating as designed.
Related NHI Mgmt Group resources: Agentic AI Identity Maturity Model · Identity Security Maturity Model · Security Questionnaires Guide
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Agentic AI Top 10 and OWASP Non-Human Identity Top 10 address the attack and risk surface, while NIST AI RMF and NIST CSF 2.0 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP Agentic AI Top 10 | ASI03 — Identity & Privilege Abuse | The article centres on agents acting under identities and access paths no one has inventoried. |
| Recommendation — Inventory agent identities and constrain delegated privileges before runtime access expands beyond review. | ||
| OWASP Non-Human Identity Top 10 | NHI-05 — Overprivileged NHI | AI agents and unmanaged AI tools can accumulate access beyond what the programme can justify. |
| Recommendation — Review AI-related non-human identities for excess access and remove privileges that lack a named business owner. | ||
| NIST AI RMF | GOVERN — AI Governance and Accountability | The piece is fundamentally about governance structure, ownership, and accountability for AI use. |
| Recommendation — Establish governance ownership, accountability, and documented oversight for every AI use case. | ||
| NIST CSF 2.0 | GV.OC-01 — Organisational Context | The article argues that organisations need a clear view of where AI is used before control can work. |
| Recommendation — Define the AI operating context, owners, and boundaries before applying controls or assessments. | ||
Key terms
- AI Governance: AI governance is the set of controls used to discover, classify, approve, restrict, monitor, and revoke AI-enabled access. It connects identity, data, and policy so organisations can manage what AI can reach, what it can share, and when it should be stopped.
- 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.
- Agent Identity: An agent identity is the set of attributes, credentials and permissions assigned to an autonomous software entity. It is treated as a non-human identity because it can authenticate, act on systems and accumulate access over time, which creates governance, audit and lifecycle obligations similar to other production identities.
- Runtime Governance: Runtime governance is the set of controls that verify what a system or agent is actually doing after deployment. It combines monitoring, authorization checks, and access validation so teams can detect drift, misuse, or excessive privilege in motion rather than assuming build-time policy still holds.
Deepen your knowledge
NHI governance, agentic AI identity, and machine identity lifecycle 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.
Published by the NHIMG editorial team on October 22, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org