An agent is software that can take an open-ended task, decide which tools to use, and act across multiple steps without a fixed script for every branch. In identity and security terms, the key issue is that an agent can change state in real systems, so its permissions and boundaries must be tightly governed.
Expanded Definition
An agent is more than an automated workflow. In NHI and agentic AI security, the term usually refers to software that can select actions, call tools, and persist across steps while holding some level of delegated authority. That makes an agent different from a simple script, because the agent can interpret context and then change state in connected systems.
Usage in the industry is still evolving. Some teams reserve “agent” for systems with autonomous planning and tool use, while others apply it to any software component that can initiate actions on behalf of a user or service. For governance, the practical distinction is whether the software can touch secrets, APIs, data, or infrastructure without a human approving every step. Guidance from the OWASP Agentic AI Top 10 and the NIST AI Risk Management Framework both point to the need for bounded action, traceability, and explicit control over tool access.
The most common misapplication is treating an agent like a passive model wrapper, which occurs when teams grant broad credentials because the system “only automates tasks.”
Examples and Use Cases
Implementing agents rigorously often introduces approval and telemetry overhead, requiring organisations to weigh speed and autonomy against containment and auditability.
- A code assistant agent opens a pull request, runs tests, and updates a ticket, but only within a limited repo scope and with short-lived credentials.
- A support agent answers customer requests and triggers account changes, using policy checks before any destructive action. See the Meta AI Instagram Account Takeover for why tool-enabled support paths must be tightly constrained.
- A data operations agent reconciles records across SaaS tools, but is blocked from exporting secrets or altering retention settings without escalation.
- An internal security agent triages alerts and enriches findings by querying logs, while remaining isolated from production deployment actions.
When organisations design these systems, they often map them to the same identity controls used for NHIs. The Ultimate Guide to NHIs — 2025 Outlook and Predictions is useful here because it frames governance, rotation, and offboarding as ongoing identity problems, not one-time setup tasks. For implementation patterns, the CSA MAESTRO agentic AI threat modeling framework offers a useful lens.
Why It Matters in NHI Security
An agent matters because it can act, and action is what turns identity risk into operational impact. If an agent has excess privileges, weak secret handling, or unclear boundaries, it can create exposure faster than a human operator could. NHIMG research shows that 97% of NHIs carry excessive privileges, which is directly relevant when an agent inherits permissions without tight scoping from the start.
That risk becomes concrete when agents interact with third-party APIs, CI/CD pipelines, ticketing systems, or production data stores. In practice, a poorly governed agent can delete records, exfiltrate sensitive data, or widen access through chained tool calls. The CoPhish OAuth Token Theft via Copilot Studio and Replit AI Tool Database Deletion are reminders that tool-enabled autonomy can turn a narrow mistake into a real incident.
Organisations typically encounter the need to define agent boundaries only after an agent has altered data, exposed a token, or executed an unwanted workflow, at which point the term 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, OWASP Non-Human Identity Top 10 and CSA MAESTRO address the attack and risk surface, while 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 Agentic AI Top 10 | A1 | Covers agentic app risks from tool use, autonomy, and misaligned actions. |
| OWASP Non-Human Identity Top 10 | NHI-02 | Agents often rely on secrets and delegated credentials, a core NHI risk area. |
| NIST AI RMF | Defines risk management for AI systems that make autonomous or semi-autonomous decisions. | |
| NIST Zero Trust (SP 800-207) | AC-6 | Zero Trust principles require least privilege for software entities like agents. |
| CSA MAESTRO | Provides threat modeling guidance for agentic AI systems and their control plane. |
Inventory agent purpose, assess harms, and enforce ongoing monitoring and human oversight.
Related resources from NHI Mgmt Group
Deepen Your Knowledge
Reviewed and updated by the NHIMG editorial team on August 27, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org
