AI features are functions in a service that use machine learning or large language model capabilities to generate, classify, recommend, or summarise output. When these features process identity or support data, organisations need clear disclosure of where data is sent, what model infrastructure is used, and what retention or training limits apply.
What AI Features Are
AI features are product capabilities that use machine learning or large language models to generate, classify, recommend, or summarise output. The security issue is not the label itself, but the data path behind it: where prompts and source data go, which model infrastructure processes them, and what retention, logging, and training limits apply.
How AI Features Change the Security Boundary
An AI feature often adds a new processing boundary inside a familiar SaaS workflow. That boundary can move content into a third-party model, a vendor-hosted inference service, or a separate AI subsystem with its own admins, telemetry, and data-use terms. Even when the user experience looks like a simple button, the underlying flow may include prompts, context injection, and post-processing that deserve the same scrutiny as any other externalised data transfer.
This matters because the feature may touch sensitive identity, support, or operational data even when the rest of the application is already approved. A recommendation engine, summariser, or classifier can still expose business context, leak secrets in prompts, or create policy gaps if the organisation assumes the parent service covers the AI layer by default.
Data, Model, and Retention Considerations
The core governance questions are straightforward: what data is sent, what the model provider can do with it, and how long outputs or prompts are retained. AI features should be reviewed for training opt-outs, retention limits, subprocessor use, regional processing, and whether vendor settings allow administrators to disable or scope the feature for specific users or content types.
Where the feature processes identity-related material, support tickets, customer records, or internal documents, organisations should treat the AI path as a distinct data-processing activity. That means understanding whether the feature stores prompts, whether outputs are reusable across tenants, and whether any hidden administrative defaults enable the feature more broadly than intended.
For product-led deployments, the most useful starting point is a governance view of the feature itself, not just the parent application. The Embedded AI in SaaS Guide is useful here because it focuses on the controls that matter most: admin settings, training terms, retention terms, and review of what the vendor actually switched on.
Operational and Security Implications
AI features can increase exposure in three common ways. First, they may expand the data surface by sending more content to external services. Second, they may create shadow usage when users enable built-in AI capabilities without formal review. Third, they may introduce misleading outputs that look authoritative but are incomplete, stale, or inaccurate, which is a security and operational problem when teams rely on them for decisions.
Discovery and inventory matter because many AI features are embedded inside familiar products rather than introduced as standalone tools. The Shadow AI and AI Agent Discovery Guide helps teams think about finding those features through SaaS, endpoint, cloud, and consent signals before they spread beyond approved use.
Where organisations need a broader control view, the AI Governance Platform Buyer’s Guide is relevant because it connects inventory, risk classification, and evaluation criteria into a practical governance model for AI-enabled products.
Risk and Threat Considerations
AI features can create material risk when users assume the parent SaaS agreement covers all model processing, or when sensitive content is sent to a model layer with looser retention and reuse terms than the base service. The practical danger is unreviewed data exposure, especially when features are enabled quietly across tenants, departments, or user groups.
Failure mechanism: The feature sends prompts, documents, or support data into a separate AI pipeline, where retention, training, or cross-tenant handling differs from the organisation’s expectations, creating disclosure, compliance, or confidentiality gaps.
Impact: Sensitive business data can leave controlled systems, be retained longer than intended, or be surfaced in ways that increase exposure, policy violations, or trust in unverified output.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
NIST SP 800-53 Rev 5 provides the primary governance reference for this term.
| Framework | Control / Reference | Relevance |
|---|---|---|
| NIST SP 800-53 Rev 5 | AU-2 — Event Logging | AI feature use creates audit needs for prompts, outputs, and admin actions. |
| AC-20 — Use of External Information Systems | AI features often route data to external model services and third-party processing. | |
| SA-9 — External System Services | AI features depend on vendor-hosted model infrastructure and service terms. | |
| Recommendation — Log AI feature enablement, data-access events, and administrative changes for review. Restrict and approve external AI processing paths before sensitive data is sent. Review provider terms, data handling, and monitoring obligations for AI services. | ||
Practitioner Guidance
What to watch for: Treat AI features as a separate approval decision when they change data movement, retention, or model access. A feature that looks like a convenience option may still need vendor review, admin scoping, and policy coverage before it is allowed for sensitive workflows.
Governance implication: Ownership should sit with the service owner and the data governance or security function together, because the decision is about both product capability and data handling. The key question is not whether the feature exists, but whether the organisation can explain and control its data path.
Related resources from NHI Mgmt Group
Deepen Your Knowledge
Free weekly newsletter
Subscribe to the NHI & AI Identity Journal
The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.
Bonus 33% off our NHI Course when you subscribe.
Reviewed and updated by the NHIMG editorial team on November 4, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org