A tool or system feature that can take actions with a degree of runtime independence rather than merely suggesting next steps. For governance teams, the key question is not whether the system looks intelligent, but whether it can act, with what approvals, and under whose authority.
Expanded Definition
Agentic capability is the property that turns a model-adjacent feature into an actor: it can execute tasks, invoke tools, and advance a workflow without waiting for a human at every step. In governance terms, the important distinction is not whether outputs are persuasive, but whether the system can cause change in systems of record, infrastructure, or data flows. That distinction is central in the OWASP Top 10 for Agentic Applications 2026 and aligns with risk thinking in the NIST AI Risk Management Framework.
Definitions vary across vendors, but in NHI security the term usually spans tool use, API calls, workflow triggers, file writes, ticket creation, and approval-chaining logic. A system may be “agentic” even when it appears narrow, if it can select actions at runtime rather than only recommend them. NHIMG’s OWASP NHI Top 10 shows why this matters: action authority is inseparable from identity, permissions, and secrets handling. The most common misapplication is treating a feature as passive automation when it can actually invoke privileged tools under inherited credentials.
Examples and Use Cases
Implementing agentic capability rigorously often introduces a control and latency tradeoff, because each new action path may require permission checks, logging, and human approval before execution.
- A support agent reads a ticket, queries an internal knowledge base, and closes the case only after policy checks confirm the response is low risk.
- A code assistant opens pull requests and updates dependencies, but cannot merge changes unless a separate approval workflow authorises the commit.
- A procurement agent drafts purchase requests and routes them to finance, while preventing direct spending without delegated authority.
- An incident-response agent rotates credentials and isolates workloads after detecting compromise, using scoped access rather than a broad admin token.
- NHIMG’s AI LLM hijack breach illustrates why agentic action paths must be bounded; the OWASP Agentic AI Top 10 also highlights tool misuse and excessive autonomy as recurring design risks.
Why It Matters in NHI Security
Agentic capability matters because every action the system can take becomes an identity question: which NHI is acting, what secrets it can reach, and whether its authority is still appropriate for the task. Once a system can initiate actions, compromised prompts, poisoned context, or abused tool access can translate directly into data exposure, unauthorized changes, or lateral movement. NHIMG research on Moltbook AI agent keys breach and the Ultimate Guide to NHIs shows how quickly agent authority and secret exposure converge into operational risk.
One NHIMG finding underscores the scale of the problem: 80% of organisations report their AI agents have already performed actions beyond their intended scope, including accessing unauthorised systems, inappropriately sharing sensitive data, and revealing access credentials. That is why governance must treat agentic capability as a privilege boundary, not a UX feature. The right question is whether runtime authority is bounded, auditable, and revocable, not whether the system can appear helpful. Organisations typically encounter the real impact only after an agent has already changed, exposed, or transmitted something it should not have, at which point agentic capability becomes 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 Agentic AI Top 10 and OWASP Non-Human Identity Top 10 address the attack and risk surface, while NIST AI RMF set the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| OWASP Agentic AI Top 10 | A01 | Agentic capability defines runtime action authority, a core concern in agentic application risk. |
| OWASP Non-Human Identity Top 10 | NHI-01 | Agent actions depend on non-human identities, permissions, and secret handling. |
| NIST AI RMF | AI RMF frames operational AI risk around governance, mapping, and control of system behavior. |
Classify every tool-using action path and restrict autonomous execution to approved, logged workflows.
