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.
Expanded Definition
AI features are product functions that use machine learning or large language models to generate, classify, recommend, summarise, or transform content. In NHI and identity-adjacent environments, the term matters because these features often process support tickets, access logs, chat transcripts, incident notes, or identity records that may contain secrets, tokens, or personal data.
Definitions vary across vendors because some products treat AI as a user-facing add-on, while others embed it deeply in search, workflow automation, or decision support. The operational question is not whether the feature sounds intelligent, but whether it sends data to third-party model infrastructure, retains prompts for training, or exposes identity-sensitive context beyond the original service boundary. That distinction aligns with broader governance thinking in the NIST Cybersecurity Framework 2.0, especially around asset visibility, risk management, and data protection.
The most common misapplication is assuming an AI feature is safe by default, which occurs when teams enable summarisation or classification on identity-related content without reviewing model routing, storage, and retention settings.
Examples and Use Cases
Implementing AI features rigorously often introduces data-handling friction, requiring organisations to weigh user convenience and automation against confidentiality, compliance, and prompt-governance overhead.
- An IAM help desk uses an AI feature to summarise tickets, but only after confirming that account recovery notes are not stored for model training.
- A security operations platform classifies alerts with AI while restricting the feature from ingesting raw API keys or session tokens from incident text.
- An HR service assistant drafts responses about onboarding, but redaction controls remove employee identifiers before prompts are sent to the model.
- A developer portal uses AI search over runbooks and logs, with disclosure that the underlying model runs in a separate environment and does not retain prompts.
These use cases become easier to govern when teams inspect the actual data path, not just the product label. NHIMG’s analysis of the DeepSeek breach shows how exposed data and embedded secrets can turn AI-enabled systems into a larger exposure surface. External guidance from the NIST Cybersecurity Framework 2.0 supports treating AI-enabled features as governed system components, not cosmetic enhancements.
Why It Matters in NHI Security
AI features matter in NHI security because they frequently sit on top of the same credentials, service accounts, API keys, and support workflows that attackers target first. If a feature ingests secrets or identity context, the risk is not limited to model output quality. It can include data exfiltration, prompt leakage, retention in vendor systems, and accidental exposure of privileged operational details.
NHIMG research on The State of Secrets in AppSec reported that 43% of security professionals are concerned AI systems may learn and reproduce sensitive information patterns from codebases, and that leaked secrets still take an average of 27 days to remediate. Those realities make AI feature governance a live security requirement, not a theoretical privacy issue.
Organisations typically encounter the operational impact only after a secret appears in a prompt, a model summary exposes restricted context, or an attacker abuses exposed credentials, at which point AI features become operationally unavoidable to address.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
OWASP Non-Human Identity Top 10 and OWASP Agentic AI Top 10 address the attack and risk surface, while NIST CSF 2.0, NIST AI RMF and NIST Zero Trust (SP 800-207) set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP Non-Human Identity Top 10 | NHI-04 | AI features can expose secrets, prompts, and data flows tied to NHI handling. |
| NIST CSF 2.0 | PR.DS | AI features must protect data in transit, use, and retention across service boundaries. |
| NIST AI RMF | AI RMF covers governance and risk management for AI-enabled system features. | |
| OWASP Agentic AI Top 10 | LLM-03 | AI features using LLMs can leak context, secrets, or instructions through prompts and outputs. |
| NIST Zero Trust (SP 800-207) | SC-7 | Zero trust requires inspecting and constraining AI feature access to sensitive resources. |
Assess AI feature risks, document mitigations, and monitor for harmful or unexpected outputs.
