AI governance remains immature as shadow AI outpaces controls

At a glance

What this is: Safe Security argues that most organisations still lack mature AI governance, and that the bigger failure is not policy absence but weak visibility, ownership and control decisions across AI use cases.

Why it matters: For IAM, NHI and AI governance practitioners, this matters because AI access, data handling and ownership controls now have to be governed as operational decisions, not assumed from acceptable-use language.

By the numbers:

• Only 5% of organizations say they have a mature AI governance program.

Context

AI governance is the set of decisions that determine who can use AI, what data it can touch, and what controls must exist before it goes live. The problem is not just policy quality. It is that many organisations have already deployed AI without a reliable way to inventory systems, assign ownership or prove that the intended controls are actually working.

The article’s central claim is that governance must be treated as an operational discipline, not a document exercise. That framing matters for IAM and NHI programmes because AI systems increasingly consume sensitive data, interact with third-party services and create new trust decisions that should be explicit rather than accidental.

Key questions

Q: How should organisations govern access to data used by AI systems?

A: Treat AI data access as an identity governance problem, not just a data storage problem. Define who or what can use each dataset, what purpose is allowed, and what runtime restrictions apply. Then review humans, service accounts, and AI agents separately so entitlement scope matches actual behaviour rather than a generic AI policy.

Q: What breaks when an AI inventory is only updated manually?

A: Manual inventories drift almost immediately in fast-moving AI estates. They miss duplicate agents, orphaned models and systems that were promoted after the last review cycle. The result is a record that looks authoritative during audits but no longer reflects production reality, which undermines accountability and access scoping.

Q: How do security teams know if AI governance is working?

A: Look for evidence that access decisions are reviewable, permissions are revocable, and exceptions are not becoming permanent. If the team cannot explain who owns an AI workflow, what it can reach, and when its access was last reviewed, governance is incomplete. Control maturity shows up in traceability, not adoption volume.

Q: When should organisations prioritise control enforcement over AI expansion?

A: Prioritise enforcement when AI systems can touch regulated or crown-jewel data but the organisation cannot yet prove who owns them or how their behaviour is monitored. Expansion before visibility turns governance into cleanup work later, while control-first deployment keeps risk decisions contained.

Technical breakdown

Why AI governance fails when inventory is incomplete

AI governance breaks down when the organisation cannot reliably identify which systems exist, who owns them, and what data they touch. A living inventory is more useful than a theoretically complete one because governance depends on current risk visibility, not static catalogues. In practice, incomplete inventory means policy, control design and monitoring all drift away from actual AI usage, especially when teams adopt tools outside formal approval paths. That is why shadow AI is primarily a visibility and ownership problem before it becomes a compliance problem. The control plane has to start with discovery, asset attribution and current-state classification.

Practical implication: build AI inventory and ownership as live control processes, not annual documentation tasks.

How sensitive-data controls change AI governance decisions

The article highlights a common governance question: whether people with access to PII or PHI should be blocked from using certain AI tools, allowed only approved systems, or placed behind additional controls such as enterprise browser enforcement. That is a governance decision about data exposure, not a generic policy statement. The underlying mechanism is simple: once AI can accept prompts containing sensitive information, the organisation must decide whether to prevent the interaction, constrain the environment, or approve a narrow use case. Those choices become part of the control design and risk acceptance record. See the AI Governance Guide.

Practical implication: define and document the sensitive-data decision path for each AI use case before broad access is granted.

Why shadow AI is an observability problem before it is a policy problem

Shadow AI emerges when acceptable-use policy exists but the organisation cannot see actual AI activity well enough to enforce it. That makes the control gap observability, not intent. Effective governance therefore needs reconciliation between intended use and observed behaviour, plus monitoring that can detect combinations of weak controls that only become dangerous when they interact. This is especially relevant where AI tools are integrated into business workflows and can touch regulated or crown-jewel data without a clear owner reviewing the risk posture. Governance matures when it can measure what is really happening, not just what was approved. See the AI Risk Assessment Guide.

Practical implication: compare declared AI use with observed AI activity and treat mismatches as governance defects.

NHI Mgmt Group analysis

AI governance is becoming a control architecture problem, not a policy-writing exercise. The article shows that organisations are being asked to orchestrate AI while still lacking the inventory, ownership and decision records needed to govern it. That means the meaningful unit of governance is the control decision, not the policy statement. Practitioners should treat AI governance as an operating model that must be testable, owned and continuously updated.

Shadow AI exposes the limits of document-first governance. When organisations can already buy or embed AI without reliable observability, the failure is not that a policy is missing from a folder. It is that the enterprise cannot reconcile what it intended with what people and systems actually do. AI security maturity model is useful here because the maturity gap is about discovery, ownership and reassessment, not slogans. Practitioners should measure visibility before they measure compliance.

AI governance debt is now a real programme risk. The longer organisations postpone decisions about approved tools, sensitive-data handling and ownership, the more they accumulate unmanaged exceptions that later become hard to unwind. That is especially relevant where AI touches identity, access and third-party workflows, because those controls are only as strong as the governance decisions behind them. Practitioners should reduce this debt by making each AI use case defensible at approval time.

The most useful governance signal is not volume, it is defensibility. A growing AI estate is not proof of maturity if teams cannot explain who owns each integration, what success means and why the data exposure is acceptable. That is the real standard the article points toward. Practitioners should judge AI programmes by whether every use case can be defended with current evidence, not by how many tools have been deployed.

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.
• 63% of organisations surveyed lacked AI governance policies to manage AI or prevent shadow AI, according to IBM's 2025 Cost of a Data Breach Report.
• Read next: Agentic AI Identity Risk Board Briefing

What this signals

AI governance debt: the longer teams delay decisions on ownership, sensitive-data handling and approved use cases, the more unmanaged exceptions they accumulate across the AI estate. That debt shows up later as remediation effort, audit friction and inconsistent policy enforcement.

The operating model is moving toward evidence-based governance, where teams are expected to reconcile intended AI use with observed behaviour and prove that controls are active, not just documented. Practitioners should prepare for a world where inventory accuracy and decision traceability matter more than policy volume.

A useful forward signal is whether AI oversight starts shifting from central policy teams to the teams closest to platform, infrastructure and data flow decisions, because that is where the actual control points now sit.

For practitioners

• Define ownership for every AI integration Assign a named owner for each AI use case, including who approves the integration, who monitors it, and who is accountable when the data or behaviour changes.
• Build a living AI inventory Track every AI asset that touches critical systems or sensitive data, and update the inventory as soon as discovery changes the known estate.
• Document sensitive-data access decisions Record whether users with PII or PHI access are blocked, restricted to approved systems, or governed by additional controls such as an enterprise browser.
• Separate KPIs from KRIs Use KPIs to show programme activity, such as systems inventoried or assessments completed, and KRIs to show whether AI risk is actually declining.

Key takeaways

• The article frames AI governance as an operational control problem, with inventory, ownership and data-handling decisions doing the real work.
• The strongest evidence is the combination of a 5% maturity signal and the broader claim that shadow AI and weak observability are outpacing formal governance.
• Practitioners should treat current-state visibility, explicit ownership and defensible approval decisions as the minimum standard for AI governance.

Related NHI Mgmt Group resources: AI Governance Guide · AI Risk Assessment Guide · AI Security Maturity Model · Agentic AI Identity Risk Board Briefing

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.
• 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.
• Risk Indicators: Risk indicators are observable actions or patterns that suggest a person or group may increase cyber risk. In human risk management, these indicators provide the evidence layer for prioritization, coaching, enforcement, and measurement, rather than relying on broad assumptions or one-time training results.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, agentic AI identity, machine identity security and secrets management. It helps practitioners connect identity control decisions to the broader security programme they run.