AI adoption is outpacing governance in manufacturing enterprises

At a glance

What this is: Cyberhaven says manufacturing AI adoption is fragmenting across sanctioned and unmanaged tools faster than governance and visibility can keep pace.

Why it matters: For IAM, NHI, and AI security teams, the issue is not AI volume alone but the loss of control over accounts, agents, and data paths as usage spreads.

Context

AI adoption in manufacturing is creating a governance gap, not just a tooling shift. As employees move across GenAI SaaS apps, endpoint AI applications, AI agents, and personal or unmanaged environments, the control problem becomes one of visibility, account governance, and data handling rather than simple application approval.

For identity and security teams, the relevant boundary is where sanctioned access ends and unmanaged use begins. When AI activity moves across multiple accounts and deployment models, conventional IT visibility and static policy enforcement no longer describe the real operating surface.

Key questions

Q: What breaks when AI governance is not aligned to how employees actually use AI?

A: AI governance breaks when policy is built around approved tools instead of real usage paths. Employees can move the same work across SaaS apps, endpoint tools, personal accounts, and agents, so the organisation loses visibility into who can access which data. The result is fragmented control, weaker auditability, and a larger exposure surface.

Q: Why do unmanaged AI accounts increase data exposure risk?

A: Unmanaged AI accounts remove policy continuity between the user, the device, and the destination service. Once corporate data is pasted into a personal account, the organisation may lose logging, retention, and enforcement. That makes even well-intentioned use risky because the business can no longer govern what happens to the copied content.

Q: How should security teams govern AI agents that can access enterprise systems?

A: Security teams should govern AI agents as non-human identities with explicit ownership, scoped privileges, and continuous monitoring. The control set should include inventory, task-bound credentials, audit trails, and revocation paths. If an agent can call tools or touch production systems, it belongs in the same governance model as service accounts and other machine identities.

Q: When should organisations unify IAM, AI governance, and data security controls?

A: Organisations should unify those controls as soon as AI use spans multiple accounts, unmanaged environments, or agentic workflows. At that point, separate policy teams cannot reliably answer basic questions about access, data handling, and accountability. A shared governance model is the only way to keep control boundaries intelligible.

Technical breakdown

Why AI governance breaks when usage fragments across accounts

AI governance depends on knowing which identities, accounts, and environments are authorised to interact with data. Once usage spreads across SaaS apps, local endpoint tools, and AI agents, the same person may operate through several different trust boundaries at once. That fragments policy enforcement, weakens auditability, and makes it harder to distinguish sanctioned use from unmanaged experimentation. In practice, the governance model stops matching the way work actually happens. Security teams then lose the ability to answer a basic question: which AI pathways can touch sensitive data?

Practical implication: move AI governance from app approval to identity-aware control of accounts, agents, and data access paths.

How unmanaged AI use expands the exposure surface

Unmanaged AI use matters because it creates a parallel execution layer outside standard IT oversight. Employees can paste sensitive content into personal environments, connect unmanaged accounts, or route work through tools that never entered the organisation's control model. That is especially relevant where AI agents can act across tasks or sessions, because the exposure is no longer limited to a single chat interface. The security problem becomes data movement through many small decisions rather than one obvious breach point. Traditional monitoring often sees the destination system after the exposure has already happened.

Practical implication: classify unmanaged AI paths as data-handling risks and monitor them with the same urgency as shadow SaaS use.

What AI agents change about identity and access governance

AI agents are not just another application category. They behave like runtime actors that can request access, use tools, and move information without a human manually authorising every step. That creates a new identity governance problem because entitlement review assumes access persists long enough to be reviewed, while agentic workflows may create and consume privileges inside one operational sequence. In manufacturing environments, where work often spans endpoints, production systems, and collaboration tools, the identity boundary becomes even harder to pin down. The result is a control gap between who or what is acting and who is accountable for the action.

Practical implication: govern AI agents as runtime identities with scoped access, not as ordinary software licenses.

NHI Mgmt Group analysis

AI adoption in manufacturing is now an identity-governance problem, not just an enablement problem. Once employees can shift between GenAI SaaS apps, endpoint AI tools, personal accounts, and AI agents, the security boundary follows the user instead of the platform. That breaks the assumption that a single IAM policy set can describe who may touch what. Practitioners should treat AI adoption as a control-surface expansion that demands identity-aware governance.

Shadow AI is best understood as unmanaged data movement with an identity wrapper. The issue is not simply that tools are undiscovered. It is that data leaves governed workflows and enters environments where account ownership, logging, retention, and revocation are inconsistent or absent. That creates audit gaps even when no malicious intent exists. Practitioners need to map where AI use escapes normal account lifecycle control.

Agentic AI forces a new concept: runtime identity drift. An AI agent can move through multiple tools and permissions during one task, so the effective identity in use changes faster than conventional review cycles can track. That means entitlement governance must move closer to issuance, delegation, and session control. Practitioners should reframe AI governance around the moment access is granted and used, not only the moment it is approved.

Manufacturing teams should expect AI governance to converge with data security and NHI control. The report's core lesson is that AI adoption patterns now intersect with machine identities, delegated access, and sensitive data exposure. That combination means the old separation between AI governance, IAM, and DLP is becoming artificial. Practitioners should build one policy view across users, agents, and the data they can reach.

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 Agent Identity Security Buyer's Guide

What this signals

Shadow AI is now an operating-model issue. When employees can reach AI through sanctioned platforms, endpoint tools, and unmanaged accounts, the security team no longer sees a single control surface. The programme response has to combine identity governance, data controls, and account visibility, or policy will continue to lag usage.

Runtime identity drift: AI agents can acquire and release permissions inside a single workflow, which means conventional access review loses its value unless issuance, delegation, and revocation are governed at the point of use. Manufacturing teams should assume that governance must shift closer to the session boundary rather than the quarterly review cycle.

For practitioners

• Map every AI access path Inventory GenAI SaaS apps, endpoint AI applications, personal accounts, and AI agents that can reach company data. Classify which identities, tokens, and sessions they use so governance covers the real pathways instead of the approved-tool list.
• Apply policy to unmanaged AI use Define which data types may never be pasted into personal or unmanaged environments and enforce that boundary with DLP, browser, and endpoint controls where possible. The goal is to reduce the number of places sensitive data can leave governed workflows.
• Treat AI agents as runtime identities Assign owners, scopes, and revocation criteria to AI agents before they are allowed to act on behalf of users or teams. Review how delegation, tool access, and session duration interact so agent behaviour stays inside a named control boundary.
• Unify identity and data governance Bring IAM, AI governance, and data security teams into one operating model for approvals, logging, and exception handling. Manufacturing environments need a shared view of which accounts and tools can touch operational data, not separate spreadsheets.

Key takeaways

• AI adoption in manufacturing is widening the gap between where work happens and where governance can actually see it.
• The report shows AI use spreading across GenAI SaaS apps, endpoint AI applications, AI agents, and unmanaged environments, which fragments control.
• Manufacturing security teams need one operating model for identities, data paths, and delegated AI access if they want policy to remain enforceable.

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.
• Runtime Identity: Runtime identity is the practice of making identity and authorization decisions at the moment an action occurs. For agents and workloads, it means access is validated against live context, not only against the identity state set during onboarding or provisioning. That makes accountability and scope enforcement possible inside fast-moving workflows.
• Agentic AI: Autonomous AI systems capable of planning, deciding, and taking actions, including calling APIs, writing code, and orchestrating other agents, with minimal human oversight. Agentic AI introduces new NHI risks as agents must authenticate to external services.
• Unmanaged AI Environment: An unmanaged AI environment is any AI tool, account, or workflow that sits outside formal enterprise control. This includes personal accounts, shadow IT, or unsanctioned endpoints. The security challenge is limited visibility into data movement, configuration, and access, which makes policy enforcement and incident response much harder.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, and secrets management in a way that fits real operating models. It helps security and identity practitioners connect access control to the broader governance issues their programmes already manage.