Agentic computing is a model where AI agents reason, decide, and act on behalf of humans across systems and workflows. In identity terms, it introduces autonomous software actors that need authentication, authorization, and audit controls comparable to human users, but with tighter runtime governance because their actions can change dynamically.
Expanded Definition
Agentic computing describes systems where AI agents can plan, invoke tools, exchange context, and complete tasks with execution authority rather than simply generating text. In NHI security, the critical shift is that the agent becomes an autonomous software actor whose identity posture must be governed like a privileged workload, but with additional runtime constraints because intent, tool use, and scope can change mid-session.
Definitions vary across vendors, but the core distinction is consistent: agentic computing is not just conversational AI, and it is not the same as workflow automation with fixed rules. It depends on an identity layer for authentication, delegated authorization, token handling, and auditability across systems. That is why frameworks such as the OWASP Agentic AI Top 10 and NIST AI Risk Management Framework are increasingly used to translate autonomy into governance controls.
The most common misapplication is treating an agent as a static service account, which occurs when teams ignore dynamic tool access, context shifts, and action-level accountability.
Examples and Use Cases
Implementing agentic computing rigorously often introduces tighter approval and logging requirements, requiring organisations to weigh automation speed against the loss of broad, implicit trust.
- An internal support agent opens tickets, queries CRM records, and drafts responses, but its permissions are constrained to specific datasets and it must log every record it touches.
- A coding agent generates and submits pull requests, while a policy engine blocks secret exposure and limits repository reach, similar to incidents discussed in NHIMG coverage such as Amazon Q AI Coding Agent Compromised.
- A finance operations agent reconciles invoices and triggers payment workflows, but high-risk actions require human approval and step-up verification before release.
- A research agent aggregates web and SaaS data, yet its access is scoped to read-only sources and monitored for prompt injection and overreach, as seen in Gemini AI Breach – Google Calendar Prompt Injection.
- Security teams model agent trust boundaries using the CSA MAESTRO agentic AI threat modeling framework and compare implementation behavior against the OWASP NHI Top 10.
Why It Matters in NHI Security
Agentic computing changes the threat model because compromise no longer requires a human to click, approve, or paste secrets. Once an agent has access to tokens, APIs, or delegated credentials, attackers can abuse its execution path to move faster than traditional monitoring and to hide within legitimate business workflows. NHIMG research on AI Agents: The New Attack Surface report shows that only 52% of companies can track and audit the data their AI agents access, leaving 48% with a compliance and investigation blind spot.
This matters because runtime failures often begin as governance failures. The same research reports that 80% of organisations say AI agents have already performed actions beyond their intended scope, including accessing unauthorised systems, sharing sensitive data, and revealing credentials. That is why agent identity, scoped authorization, token lifecycle management, and immutable audit trails must be treated as core controls, not optional enhancements. Relevant guidance also appears in the NIST AI Risk Management Framework and the MITRE ATLAS adversarial AI threat matrix.
Organisations typically encounter the operational cost of agentic computing only after an agent has overreached, leaked data, or triggered an unauthorized action, at which point the need for identity governance 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 Non-Human Identity Top 10 and OWASP Agentic AI Top 10 address the attack and risk surface, while NIST AI RMF, NIST CSF 2.0 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-02 | Agentic systems depend on secure secret and token handling for autonomous actions. |
| OWASP Agentic AI Top 10 | A1 | Autonomous action and tool misuse are central risks in agentic applications. |
| NIST AI RMF | Defines risk governance practices for AI systems, including autonomy and accountability. | |
| NIST CSF 2.0 | PR.AA-01 | Identity and access assurance apply directly to non-human actors with execution authority. |
| NIST Zero Trust (SP 800-207) | SC-7 | Agent traffic and permissions should be constrained under zero trust segmentation principles. |
Inventory, rotate, and scope all agent credentials so tool use cannot exceed intended authority.
