Data and AI security depends on classification and access governance

At a glance

What this is: NHI Mgmt Group's analysis, drawing on Commvault's blog post: Data and AI security now depends on discovery, classification, and access governance across users, systems, and AI workflows, with visibility into sensitive data treated as the first control that makes everything else workable.

Why it matters: IAM, PAM, and NHI teams need this lens because overpermissive human and machine access only becomes governable when sensitive data can be located, classified, and tied to policy.

By the numbers:

• 90% of organizations have exposed sensitive cloud data that can be surfaced by AI, according to Commvault.
• 40% of files uploaded into generative AI tools contain Personal Identifying Information or PCI data, according to Commvault.

Context

Data security is no longer just about keeping attackers out of a perimeter. In cloud and AI-heavy environments, sensitive information is spread across databases, files, SaaS tools, endpoints, and model workflows, which means security teams must first know where data lives before they can govern it effectively.

The identity angle is increasingly direct because access to data is now shared by users, applications, service accounts, and AI systems. When those identities retain broad access after their business need has passed, classification and policy enforcement become the only practical way to reduce exposure at scale.

Key questions

Q: How should security teams govern sensitive data used by AI systems?

A: Security teams should treat AI as a data consumer that needs policy boundaries, not just authentication. Classify sensitive data, define which datasets may enter AI workflows, and monitor outputs, logs, and downstream reuse. If governance stops at login, the organisation can approve access while still losing control of the data itself.

Q: Why do over-permissive service accounts increase risk in regulated environments?

A: Over-permissive service accounts turn a single machine identity into a broad trust anchor across multiple systems. In regulated environments, that creates both security exposure and evidence problems, because it becomes difficult to justify why the account still needs every privilege it was ever given.

Q: What breaks when organisations do not know where sensitive data is stored?

A: Identity controls lose their target. If data location is unknown, then access review, audit evidence, and response prioritisation all become weaker because security teams cannot connect identities to the repositories they actually touch. In practice, unknown data usually means unknown exposure.

Q: How should organisations respond when sensitive data starts flowing into AI pipelines?

A: Treat AI pipeline exposure as a governance boundary change, not just a storage issue. Reclassify the risk, check whether the data can be accessed by copilots or automation, and tighten policy where necessary. The key is to control movement before the pipeline turns sensitive data into operational input.

Technical breakdown

Why data discovery is now the first control for sensitive data

Discovery is the process of locating sensitive data across structured, semi-structured, and unstructured environments. In modern estates, that includes cloud storage, SaaS applications, endpoints, and duplicated content that may sit outside normal application ownership. Without discovery, classification and access governance are always incomplete because control policy is applied to only part of the real data estate. The technical problem is not just volume, but fragmentation: data moves faster than manual inventory and often exists in shadow locations that standard controls never see.

Practical implication: teams should treat incomplete data discovery as a governance gap, not a reporting gap.

How classification turns raw data into enforceable policy

Classification assigns meaning to data so policy can follow sensitivity rather than storage location alone. A file or record marked as PII, PCI, PHI, intellectual property, or secrets can then be handled differently for retention, masking, monitoring, and access restriction. This matters because controls based only on system boundaries ignore the actual risk of the content. In AI workflows, classification also has to apply before training and before runtime use, otherwise regulated or sensitive information can enter model inputs and outputs without any policy decision.

Practical implication: classify before training or broad distribution so downstream controls can enforce sensitivity-based handling.

Why access governance must include service accounts and AI systems

Access governance answers who or what can use data, under what conditions, and with what level of privilege. That scope must include human users, applications, service accounts, and AI systems because non-human identities now account for a large share of data access activity. The risk is overpermissive access, where identities keep broad entitlements long after their operational need changes. In practice, this creates a larger attack surface, increases insider and breach blast radius, and makes compliance claims hard to defend because the policy is not aligned to actual use.

Practical implication: extend entitlement review and drift detection to machine identities, not just employees.

Threat narrative

Attacker objective: The objective is to reach sensitive data through weak visibility and overpermissive access, then use or expose it in ways that evade governance and increase regulatory and operational risk.

1. Entry occurs when sensitive data is present in cloud systems, shared files, or AI inputs without reliable discovery and classification.
2. Credential and access abuse follows when users, applications, or service accounts retain permissions that exceed their data-handling need.
3. Impact emerges through accidental exposure, model leakage, insider misuse, or compliance failure when regulated data is processed without policy controls.

Breaches seen in the wild

• 12,000 secrets in LLM training data: Truffle Security found 11,908 live API keys and passwords hard-coded in web pages captured by Common Crawl, a dataset used to train LLMs.
• DeepSeek database exposure 2025: An unauthenticated DeepSeek ClickHouse database exposed over a million log lines with plaintext chat history and API keys in 2025.

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

Data visibility is now the control boundary, not a supporting control. Perimeter tools cannot govern what they cannot find, and that is why discovery and classification sit ahead of access policy in any serious data programme. In cloud and AI estates, this becomes a governance issue as much as a security one. Practitioners should treat incomplete discovery as a direct limitation on control effectiveness, not as an acceptable coverage gap.

Overpermissive access is the data security equivalent of standing privilege. The article’s most important identity signal is that users, applications, service accounts, and AI systems can all hold unnecessary rights to sensitive data. That makes data classification and entitlement governance inseparable. When access is not tied to sensitivity, the organisation cannot defend who should be able to see regulated data, so the control model has to shift from static permission grants to context-aware policy.

AI governance debt accumulates when training data is not classified before use. The article correctly frames model risk as a data problem first and an AI problem second. If sensitive or regulated information enters training sets or runtime prompts without classification, the model inherits the exposure. The named concept here is classification drift: the gap between where data sensitivity changes and where policy still assumes the old label. Practitioners should close that gap before AI rollout scales it.

Compliance evidence depends on data lineage, not policy intent. GDPR, HIPAA, and PCI DSS obligations become defensible only when organisations can show where sensitive data resides, who accessed it, and under what governance. That is an operational evidence problem, not just a legal one. For identity teams, the practical conclusion is that access governance and audit evidence must be built from the same data inventory, or compliance claims will remain fragile.

From our research library:

• Only 23% of IT leaders were very confident in their organisation's ability to manage security and governance for GenAI deployments, according to a 2025 Gartner survey of 360 IT leaders.
• Read next: AI Security Platform Buyer's Guide

What this signals

Classification drift: the real governance gap is that data sensitivity changes faster than policy models do. Once sensitive data has spread into shared files, cloud services, and AI workflows, security teams need discovery and classification to close the gap before access control can matter.

The identity problem is not only human access. Service accounts and AI-enabled workflows expand the number of principals touching sensitive data, so entitlement review has to include machine identities and not stop at employee accounts. In practice, that means access governance must be tied to data sensitivity, not just directory structure.

AI programmes create a new exposure path when training or prompt data contains regulated content. The practical signal for practitioners is simple: if data classification is not part of AI intake, then compliance, masking, and access decisions are all being made too late.

For practitioners

• Implement enterprise-wide sensitive data discovery Inventory structured, semi-structured, and unstructured data across cloud platforms, SaaS applications, endpoints, and shared repositories before expanding AI use cases.
• Classify data before AI training or prompt ingestion Block or redact PII, PCI, PHI, secrets, and other regulated content before it enters training sets, embeddings, or runtime prompts.
• Extend access reviews to service accounts and AI systems Review non-human identities alongside users so overpermissive access, drift, and stale entitlements are captured in the same governance cycle.
• Tie retention, masking, and deletion to classification labels Use sensitivity labels to drive retention, masking, redaction, and deletion policies so controls follow the data rather than the storage layer.

Key takeaways

• The article’s central risk is not data volume alone, but the inability to find, classify, and govern sensitive information before it spreads across cloud and AI workflows.
• Identity and access control become materially weaker when service accounts, applications, and AI systems retain broad permissions to data that has not been mapped to sensitivity.
• The most effective control shift is to make discovery and classification the input to access governance, retention, masking, and audit evidence rather than treating them as separate programmes.

Key terms

• Data Discovery: Data discovery is the process of finding where information lives across cloud, SaaS, endpoints, backups, and analytics systems. In practice, it creates the inventory that makes classification, access decisions, recovery planning, and AI governance possible rather than speculative.
• Data classification: Data classification is the process of labelling information according to sensitivity, regulatory impact, or business value so controls can be applied consistently. For AI governance, it allows policy to follow the data into prompts, sessions, and destinations rather than relying on brittle text matching.
• Access Governance: Access governance is the policy and workflow layer that manages how access is requested, approved, certified, and revoked. In SaaS environments it helps standardise control across many applications, reducing inconsistency between teams. It is most effective when it covers both human accounts and non-human identities.
• Over-Permissive Access: Over-permissive access means a credential can reach more data or systems than the task requires. In non-human identity environments, this often happens when apps are granted broad read, write, or delete rights, or when permissions are difficult to interpret. It increases blast radius and makes compromise more damaging.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, and secrets management. It helps practitioners connect data access decisions to the identity controls their programmes already depend on.