Security teams should start by identifying where AI systems are in use, what data they touch, and which identities can reach them. That discovery phase creates the control map for policy, classification, access restriction, and monitoring. Without it, organisations tend to protect known assets while leaving AI connected systems and data flows outside governance.
Why This Matters for Security Teams
AI discovery is the point where security teams stop guessing and start controlling. If an organisation cannot identify which models, copilots, pipelines, agents, and data stores are in use, broader data protection efforts become partial at best. Discovery exposes where sensitive data is already flowing, which identities can reach it, and which systems create the highest-risk exposure paths. That makes it the prerequisite for classification, policy scoping, and access tightening.
This sequencing aligns with the NIST Cybersecurity Framework 2.0, which treats asset visibility as a foundation for risk management rather than an afterthought. NHIMG research points to why that matters: in the State of Non-Human Identity Security, only 1.5 out of 10 organisations reported high confidence in securing NHIs, while 85% lacked full visibility into third-party vendors connected via OAuth apps. In practice, many security teams encounter data leakage and overexposure only after an AI-enabled workflow has already been deployed into production.
How It Works in Practice
Effective sequencing usually starts with inventory, then moves to control design. The first task is to identify every AI touchpoint, including public models, internal copilots, agentic workflows, embedded AI in SaaS, and any service accounts or secrets that let those systems operate. Security teams should also map the data classes involved, such as source code, customer records, regulated data, prompt inputs, retrieved context, and output destinations.
From there, discovery should produce a control map that links AI use cases to specific protection steps. That map can then drive policy-as-code, access review, and logging requirements. For example:
- Classify which datasets may be used for training, retrieval, or prompt enrichment.
- Identify which human and non-human identities can call models, tools, and plugins.
- Flag where secrets, API keys, or tokens are exposed inside AI pipelines.
- Apply monitoring to high-risk flows before expanding to full-scale data loss prevention.
This is also where the NHI Lifecycle Management Guide becomes operationally useful, because AI systems often depend on short-lived credentials, scoped tokens, and service identities that must be discovered before they can be governed. Current guidance suggests combining discovery with access baselining rather than waiting for a perfect catalog. The practical goal is not complete certainty, but a defensible map of where data actually moves and which identities can move it.
That sequencing also helps teams apply mature control families such as CIS Controls v8 and data protection obligations under the EU General Data Protection Regulation (GDPR) in the right order. These controls tend to break down when AI use is embedded in shadow IT workflows, because the underlying models, connectors, and identities are not visible to the team building the protection plan.
Common Variations and Edge Cases
Tighter discovery often increases operational overhead, requiring organisations to balance visibility against deployment speed. That tradeoff is real, especially when AI is already embedded in developer tools, customer service platforms, or third-party SaaS integrations. Best practice is evolving, but there is no universal standard for perfect AI asset discovery yet, so teams should prioritise high-risk systems first.
One common edge case is agentic AI, where a single system can chain tools, call APIs, and move data across multiple environments. In those environments, discovery should include not only the model endpoint but also the agent’s execution rights, downstream connectors, and the identities used to invoke each step. Another edge case is regulated data: if a system may touch payment, health, or personal data, the discovery phase should immediately trigger stricter scoping and retention review rather than a broad exploratory rollout.
NHIMG’s Top 10 NHI Issues and Ultimate Guide to NHIs — Key Challenges and Risks both reinforce the same operational point: discovery is not a paperwork exercise, it is the mechanism that prevents invisible identities and hidden data paths from undermining the rest of the program. In mixed environments, discovery-first programs still struggle when business units can create new AI services without central registration, because the inventory becomes stale before controls are applied.
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 CSA MAESTRO 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 |
|---|---|---|
| NIST CSF 2.0 | ID.AM | Asset management is the prerequisite for AI discovery and control scoping. |
| OWASP Non-Human Identity Top 10 | NHI-01 | Discovery must identify non-human identities and their reachable assets. |
| CSA MAESTRO | GOV-1 | Governance starts with knowing where agentic AI is deployed and who controls it. |
| NIST AI RMF | GOVERN | The govern function depends on visibility into AI use and data exposure. |
| NIST Zero Trust (SP 800-207) | PR.AC | Zero trust requires knowing which identities and systems need access. |
Inventory AI systems, identities, and data flows before assigning protection controls.
Related resources from NHI Mgmt Group
- How should security teams govern data protection when AI adoption expands across enterprise systems and compliance obligations increase?
- How should security teams prepare data governance programs for fast-moving AI, privacy, and cyber regulations?
- How should security teams extend data protection to AI interactions without replacing existing controls?
- How should security teams keep SaaS application data accurate across discovery, mapping, and reporting?
