When AI vendors, embedded AI features, and agent connectors are missing from supplier and access governance, organisations lose control over who can process data and under what terms. That usually leads to shadow AI, excessive permissions, and unknown data exposure. It also weakens risk assessment, because the organisation cannot prove which tools touched which information.
Why This Matters for Security Teams
When AI usage is omitted from supplier and access controls, the organisation still has a control framework on paper but not in practice. AI vendors can receive data, embedded AI features can make decisions, and agent connectors can act with privileges that were never explicitly approved. That creates a gap between procurement, identity governance, and data protection, which is exactly where shadow AI tends to grow.
The main issue is not only exposure, but loss of accountability. If an AI tool can read tickets, documents, customer records, or source code, security teams need to know the legal basis, retention terms, subprocessors, and identity path behind that access. Guidance from OWASP Non-Human Identity Top 10 is especially relevant here because many AI services and agents behave like non-human identities even when procurement treats them as ordinary software. In practice, many security teams encounter this only after a data-flow review, incident, or audit finding has already exposed the missing control boundary.
How It Works in Practice
In mature environments, supplier and access governance should treat AI as a distinct class of third party and a distinct access actor. That means identifying whether the AI is a hosted SaaS feature, an embedded function in another product, a model endpoint, or an autonomous agent with tool access. Each has different approval, monitoring, and revocation requirements. Current practice is to align this with vendor risk management, identity governance, and data classification so that access is approved based on what the AI can reach, not just who bought the product.
A practical control set usually includes:
- Listing every AI supplier, embedded AI feature, and agent connector in the approved vendor register.
- Mapping each AI tool to the data types it can process, store, or transmit.
- Assigning a named business owner and technical owner for each AI-enabled service.
- Using least privilege for API keys, service accounts, and delegated tokens.
- Rechecking contracts for subprocessors, retention, training use, and breach notification terms.
- Logging prompts, outputs, and tool calls where this is proportionate and legally permitted.
For control design, NIST SP 800-53 Rev 5 Security and Privacy Controls provides a useful structure for supplier, access, audit, and data protection expectations, while CIS Controls v8 helps operationalise inventory, access management, and configuration discipline. Organisations handling payment data should also check PCI DSS v4.0 where AI touches cardholder data or connected payment workflows. These controls tend to break down in fast-moving SaaS environments where product teams can enable AI features without security review because the procurement record does not capture feature-level access.
Common Variations and Edge Cases
Tighter AI supplier control often increases operational overhead, requiring organisations to balance speed of adoption against assurance, especially when business units expect immediate access to new AI features. Best practice is evolving for embedded AI, so there is no universal standard for how deeply every feature must be assessed, but the governance principle is clear: if the feature can process organisational data or act on behalf of a user, it belongs in scope.
Edge cases appear when the AI is bundled into a platform that the organisation already trusts, such as productivity suites, CRM tools, or developer platforms. The risk is that teams assume the base vendor approval covers the AI capability, when in reality the AI may introduce new data processing, model training, or agentic access paths. Another common blind spot is delegated access through a human account. That can look compliant in an access review while still giving an AI connector broad reach that no one intended. NHI Management Group recommends treating these connectors as controlled non-human identities whenever they hold tokens, refresh credentials, or execute actions independently.
For governance-heavy programmes, ISO/IEC 27001:2022 Information Security Management is useful for anchoring supplier control, access review, and continual improvement expectations. The practical test is simple: if the organisation cannot say which AI tool touched which information, under whose authority, and under which contract terms, then the control model has already failed.
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 address the attack and risk surface, while NIST CSF 2.0, NIST SP 800-53 Rev 5 and CIS Controls v8 set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP Non-Human Identity Top 10 | AI services and agents often behave like non-human identities with their own credentials and privileges. | |
| NIST CSF 2.0 | GV.SC-1 | Supplier governance is needed to track third-party AI and embedded features. |
| NIST SP 800-53 Rev 5 | SA-9 | External system services controls cover supplier-provided AI processing and dependencies. |
| CIS Controls v8 | 5 | Account and access management is central when AI tools can act on behalf of users. |
Inventory AI tools as identities, control their secrets, and revoke access when usage is no longer approved.
Related resources from NHI Mgmt Group
- What breaks when AI privacy controls are used as a substitute for access governance?
- What breaks when AI models can access sensitive data without output controls?
- What breaks when AI systems can access data without context-aware controls?
- What breaks when access controls do not account for AI correlation risk?
Deepen Your Knowledge
Reviewed and updated by the NHIMG editorial team on August 24, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org
