An AI model tuned to one organisation’s environment rather than a shared cross-customer dataset. In identity security, this usually means the model can use local signals to improve recommendations while raising specific questions about isolation, residency, retention, and whether customer data is reused for training.
Expanded Definition
A customer-specific model is tuned to one organisation’s environment, so its outputs reflect local signals, policies, and workflows rather than a shared cross-customer baseline. In NHI security, that distinction matters because the model may process identity telemetry, access patterns, secret usage, or remediation history that is sensitive in its own right.
Definitions vary across vendors on how much customer data can be retained, reused, or blended into future training. A stricter reading treats the customer boundary as part of the control plane, not just a billing boundary. That means practitioners should ask whether the model is isolated at inference time, whether prompts and outputs are retained, and whether customer data can improve other customers’ models. For governance, the most useful reference points are NIST Cybersecurity Framework 2.0 and the NHI lifecycle and visibility guidance in Ultimate Guide to NHIs.
The most common misapplication is assuming “customer-specific” automatically means isolated and non-reusable, which occurs when retention and training terms are left implicit in the deployment contract.
Examples and Use Cases
Implementing a customer-specific model rigorously often introduces data-governance and residency constraints, requiring organisations to weigh better recommendations against tighter controls on what the model can store, infer, or export.
- An NHI platform tunes anomaly detection to one tenant’s service-account usage so it can distinguish normal automation from credential abuse.
- A secrets management workflow adapts remediation suggestions to a customer’s vault layout, rotation cadence, and approval paths, as discussed in the Ultimate Guide to NHIs.
- A SOC assistant learns local IAM naming conventions and environment tags so it can classify alerts more accurately without relying on a generic model.
- A compliance team limits model training to approved regional data because the customer’s data residency obligations prevent cross-border reuse.
- An enterprise requires an external assurance review against NIST Cybersecurity Framework 2.0 before any tenant-specific tuning is activated.
Why It Matters in NHI Security
Customer-specific models can improve precision, but they also concentrate risk if the tuning set contains secrets, privileged access patterns, or incident data that should never leave the tenant boundary. That is especially important in NHI programs, where Ultimate Guide to NHIs reports that 97% of NHIs carry excessive privileges and only 5.7% of organisations have full visibility into their service accounts. If a model is trained on incomplete or overexposed identity data, it may reinforce unsafe recommendations rather than correct them.
From a governance perspective, the central question is whether the model is simply personalised or whether it becomes a new repository of customer evidence. That difference affects retention, deletion, auditability, incident response, and contract language. It also changes how teams evaluate third-party tooling under the NIST Cybersecurity Framework 2.0. Organisations typically encounter the operational cost of a customer-specific model only after a data-handling review, breach inquiry, or export-control issue forces them to prove what the model saw and what it retained.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Agentic AI Top 10 address the attack and risk surface, while NIST AI RMF and NIST CSF 2.0 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP Agentic AI Top 10 | Covers agentic model data handling and tenant-specific prompt risks. | |
| NIST AI RMF | Addresses AI governance, data lineage, and model risk controls for tuned models. | |
| NIST CSF 2.0 | PR.DS-1 | Data-at-rest protections apply to customer-specific training and inference data. |
Document training data scope, retention limits, and validation steps for each customer-specific model.
