TL;DR: McKinsey finds that only 7% of organizations have successfully scaled AI across the enterprise, reinforcing that trusted, governed, AI-ready data matters more than model choice or infrastructure, according to Sentra. The governance shift is now about visibility, metadata, access context, and continuous control across the data AI can retrieve and transform.
At a glance
What this is: This analysis argues that enterprise AI stalls when data is not discoverable, classified, governed, and usable by AI systems at scale.
Why it matters: It matters because security, data, and IAM teams now have to govern what AI can retrieve, combine, and expose, not just who can open a file.
By the numbers:
- Only 7% of organisations have successfully scaled AI across the enterprise, according to McKinsey research cited by Sentra.
Context
The governance gap is no longer about whether an organisation can deploy AI tools. It is about whether the underlying data is discoverable, classified, trusted, and available to AI systems in a controlled way. When that foundation is weak, AI scales confusion as quickly as it scales automation.
AI changes the security model because it does not consume data like a traditional application. Large language models, copilots, retrieval-augmented generation systems, and agents can pull from multiple repositories, combine content, and generate new outputs, which makes data governance an access-control problem as much as a content-quality problem.
For identity and security teams, the key shift is that access context now matters at the point of retrieval and use, not only at the point of storage. That makes the article relevant to IAM, NHI governance, and broader data security programmes because AI can amplify overly broad permissions and stale data exposure.
Key questions
Q: How should security teams govern personal data used by AI agents?
A: Security teams should govern agent access as a runtime control problem, not as a one-time permission decision. Limit the data each agent can reach, bind access to a specific task, and monitor actual behaviour continuously. That approach makes privacy records and IAM controls reflect the same operational reality.
Q: Why does poor metadata create risk for AI systems even when the model is strong?
A: Because the model only reasons over what retrieval gives it. If the retrieved content is stale, misclassified or stripped of context, the output can be confidently wrong while still appearing authoritative. The failure is upstream in the data foundation, which means model upgrades alone do not solve governance, compliance or accuracy problems.
Q: What breaks when organisations rely on quarterly access reviews?
A: Quarterly reviews break the link between policy and reality. Access can drift, accounts can become orphaned, and toxic combinations can appear and disappear between review cycles. By the time the certification happens, the control is describing a past condition rather than the current risk posture.
Q: What should data and security teams align on before scaling AI?
A: They should align on one operating model for discovery, classification, access, lineage, and remediation. AI readiness fails when security owns permissions but data teams own structure and quality in separate workflows. Shared governance is what turns enterprise data into something AI can use safely and consistently.
Technical breakdown
Why AI-ready data changes the governance model
AI-ready data is not just well-organised data. It is data that can be discovered, classified, traced, and used with enough context for machine systems to make reliable decisions. Traditional controls assumed a person or application would retrieve a fixed record and stop there. AI systems instead retrieve, merge, summarise, and regenerate content across multiple sources, which changes the control point from storage alone to the retrieval and transformation path. That is why metadata, lineage, and sensitivity labels become operational controls rather than administrative extras.
Practical implication: govern the retrieval path, not only the repository.
Why metadata and lineage are now security controls
Metadata gives AI the context needed to distinguish public, internal, regulated, and highly sensitive information. Lineage shows where data came from, how it moved, and whether it is safe to reuse in a downstream model response. Without those signals, AI can treat stale, duplicated, or low-trust content as authoritative. In practice, that creates a governance problem that looks like an access problem, a data quality problem, and a compliance problem at the same time. The control objective is to make AI decisions explainable enough to trust and restrictive enough to govern.
Practical implication: treat classification and lineage as inputs to policy enforcement.
How continuous governance supports AI scale
Periodic access reviews and one-time classification projects do not match the pace of AI usage. AI systems operate continuously, and the data landscape changes continuously too, with new repositories, stale files, overexposed records, and changing entitlements. Continuous governance means discovery, classification, monitoring, and remediation run as an ongoing control loop rather than a quarterly checkpoint. That matters because AI will reuse whatever it can reach, including redundant or obsolete data that humans might have ignored. The governance model has to keep pace with machine-scale consumption.
Practical implication: move from periodic review to continuous data control.
NHI Mgmt Group analysis
AI data readiness is becoming an identity governance problem as much as a data governance problem. The article makes clear that AI value depends on who and what can reach enterprise data, not just where the data sits. Once copilots and agents enter the picture, access scope, ownership, and lifecycle context become part of AI trust. Practitioners should treat AI data readiness as a shared IAM and data governance discipline, not a storage hygiene exercise.
Metadata is the missing control plane for enterprise AI. Classification, ownership, sensitivity, and lineage are what allow policy decisions to follow the data into AI workflows. Without those signals, organisations are left trying to govern outputs after the fact, which is too late for meaningful control. The practical conclusion is that AI governance starts with machine-readable context, not with model tuning.
Continuous governance replaces the old review cycle. Quarterly access reviews and annual audits were built for slower-moving information estates. AI retrieves and recombines data at a pace that makes static governance stale almost immediately. Organisations that rely on periodic review alone will miss the overexposed, duplicated, and obsolete content that AI is most likely to surface. Practitioners should rebuild control monitoring around continuous discovery and remediation.
AI scale will widen the gap between organisations that know their data and those that only think they do. The article signals that enterprise AI maturity will increasingly depend on operational visibility, not experimentation volume. That changes the competitive baseline for security and data teams alike. The organisations that can prove what AI can access, why it can access it, and whether that access remains appropriate will be the ones able to scale responsibly.
From our research library:
- Enterprise AI use rose from 55% of organisations in 2023 to 88% in 2025, according to McKinsey’s Global Surveys on the State of AI.
What this signals
AI data readiness is now a governance prerequisite, not a downstream optimisation. Security programmes that wait until after AI rollout to fix access context and metadata will be governing at the wrong layer. The right control point is the dataset AI can retrieve and transform, because that is where trust and exposure are created.
Metadata becomes the mechanism that lets policy travel with the data. In practice, classification and lineage are what stop AI from collapsing public, regulated, and sensitive content into one undifferentiated response stream. That is why AI governance, data security, and IAM now intersect in the same operational workflow.
Continuous monitoring matters more than periodic certification for AI workloads. AI systems do not wait for quarterly review cycles, so stale data and overexposed access remain live governance problems until they are continuously detected and remediated. Programmes that keep using legacy review cadences will struggle to prove that AI access remains appropriate over time.
For practitioners
- Implement continuous data discovery Map where sensitive data lives across cloud, SaaS, on-premises, and AI data stores so AI access decisions are based on current inventory rather than assumptions.
- Attach metadata to AI-accessible data Enforce classification, ownership, lineage, and sensitivity labels so AI systems can distinguish public content from regulated or highly sensitive material.
- Right-size access before AI expansion Review who and what can reach high-value datasets, including copilots and agents, and remove overexposed entitlements that would otherwise scale with AI usage.
- Replace periodic review with continuous remediation Move governance from quarterly checks to ongoing monitoring that can detect stale, duplicate, or obsolete data before AI amplifies it.
- Build a shared operating model between security and data teams Define one governance workflow for identity, access, data quality, and lifecycle decisions so AI readiness is not split across disconnected functions.
Key takeaways
- AI fails at enterprise scale when the data it reaches is not discoverable, classified, and governed in a way the model can trust.
- Traditional document-level controls are not enough because AI retrieves, combines, and generates information across multiple systems in real time.
- Security teams and data teams now need a shared control model for visibility, metadata, lineage, and continuous remediation.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
MITRE ATT&CK and OWASP API Security Top 10 address the attack and risk surface, while NIST AI RMF, NIST CSF 2.0 and NIST SP 800-53 Rev 5 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST AI RMF | GOVERN — AI Governance and Accountability | The article is fundamentally about AI governance operating model and accountability. |
| Recommendation — Establish governance for AI-accessible data and assign clear accountability for discovery, access, and remediation. | ||
| NIST CSF 2.0 | PR.DS-01 — Data-at-rest is protected | AI readiness depends on controlling the data that AI can retrieve and reuse. |
| Recommendation — Protect sensitive data stores and enforce data handling rules before AI systems can consume the content. | ||
| NIST SP 800-53 Rev 5 | AC-6 — Least Privilege | The article highlights overexposed data access that AI can amplify. |
| Recommendation — Apply least-privilege access to datasets and AI-connected services to reduce unnecessary retrieval scope. | ||
| MITRE ATT&CK | TA0007;TA0009 — Discovery; Collection | AI systems discover and collect data across multiple sources, which mirrors relevant adversary behaviors. |
| Recommendation — Map AI retrieval paths to discovery and collection risks so you can monitor what content AI can aggregate. | ||
| OWASP API Security Top 10 | API8 — Security Misconfiguration | AI integrations often fail when the surrounding data and access configuration is too loose or inconsistent. |
| Recommendation — Review AI-connected APIs and data services for misconfiguration that exposes more data than intended. | ||
Key terms
- AI Data Readiness: AI Data Readiness describes whether an organisation can safely expose data to AI systems without losing control over sensitivity, purpose, or access scope. It combines discovery, permission management, and continuous oversight so data use remains aligned to governance expectations.
- Metadata Governance: Metadata governance is the discipline of defining, capturing and protecting the information that proves where a record came from and how it changed. In regulated life sciences, it turns data into evidence by making lineage, ownership, timestamps and alteration history consistently auditable across systems.
- Continuous governance: An identity governance model that checks and enforces policy as activity happens rather than on a schedule. It is designed to catch drift, misuse, and orphaned access while the identity is still active, which matters when risk unfolds in minutes instead of review cycles.
- AI retrieval surface: The set of repositories, datasets, and services an AI system can query, combine, and transform during inference. This surface becomes a governance boundary because whatever sits inside it can be surfaced in outputs, making access scope and data context central security concerns.
Deepen your knowledge
The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, agentic AI identity, machine identity security, IAM, and secrets management. It helps practitioners connect identity controls to the broader governance requirements that AI-driven programmes now depend on.
Published by the NHIMG editorial team on August 18, 2026.
Updated on October 12, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org