Data-Centric Controls

Data-centric controls are security measures that protect information directly, rather than relying only on network or perimeter barriers. They include classification, access restrictions, masking, policy enforcement, and monitoring tied to the sensitivity and location of the data. This approach is essential when AI systems can reach many data sources at once.

Expanded Definition

Data-centric controls are the measures that follow the information itself, not the network edge around it. In practice, they combine classification, encryption, masking, tokenisation, access restrictions, and policy enforcement so that sensitivity travels with the data across SaaS, cloud storage, analytics platforms, and AI workflows. This matters because identity-based access alone does not explain what should happen after an AI agent, service account, or pipeline reaches the data.

Definitions vary across vendors, especially when products blend data loss prevention, information rights management, and policy engines into one label. For NHI security, the most useful interpretation is operational: controls should be enforceable at read, write, share, export, and inference time, not just at login. That aligns with the risk-based approach described in the NIST Cybersecurity Framework 2.0 and with NHIMG guidance on data exposure patterns in the Ultimate Guide to NHIs — Standards. The most common misapplication is treating classification as the control itself, which occurs when teams tag data but do not enforce downstream policy on the systems that consume it.

Examples and Use Cases

Implementing data-centric controls rigorously often introduces friction for automation and analytics, requiring organisations to weigh broad machine access against tighter handling of sensitive records.

• An AI agent queries a customer support repository, but masked fields hide payment details unless the agent is explicitly authorised for that purpose.
• A service account can read source data in a warehouse, yet export to an external tool is blocked by policy unless the destination is approved and logged.
• A CI/CD pipeline can access secrets only for a limited deployment window, while the data store enforces encryption and field-level restrictions on retrieved configuration data.
• An internal search assistant can index documents, but classification rules prevent regulated records from being surfaced in responses to unauthorised users.
• NHIMG research shows how badly data and secret exposure compounds when controls are weak, including the finding that 79% of organisations have experienced secrets leaks, with 77% causing tangible damage, in the Ultimate Guide to NHIs — Key Research and Survey Results.

For implementation patterns, the relevant external baseline is the NIST Cybersecurity Framework 2.0, which helps translate protection goals into practical safeguards.

Why It Matters in NHI Security

NHI environments break perimeter assumptions because agents, APIs, scripts, and service accounts often touch multiple data domains in a single workflow. If data-centric controls are absent, a legitimate identity can still move sensitive information into places it should never reach, such as logs, prompts, replicas, backups, or downstream SaaS tools. That is especially dangerous when an AI system has broad retrieval capability but weak data filtering, because the identity may be valid even when the data use is not.

NHIMG research shows the scale of the problem: 96% of organisations store secrets outside secrets managers in vulnerable locations, and 73% of vaults are misconfigured, both of which amplify data exposure for non-human actors. Data-centric controls therefore become a governance requirement, not just a privacy feature, because they constrain what an NHI can do after access has already been granted. The same logic applies to least privilege and Zero Trust, where enforcement must follow the asset rather than stop at authentication. Organisations typically encounter the need for data-centric controls only after an AI agent over-shares, exfiltrates, or replicates sensitive records, at which point the term becomes operationally unavoidable to address.

Related resources from NHI Mgmt Group

• How should security teams implement data-centric controls for AI agents in enterprise environments
• Why do identity-centric attacks bypass traditional security controls so often?
• Why do non-human identities complicate data protection controls?
• When do on-prem data controls become an NHI issue?