Join our Newsletter — 33% off our NHI Course
Home› Glossary› Agentic AI & Autonomous Identity› Agentic Computing
Agentic AI & Autonomous Identity

Agentic Computing

← Back to Glossary
By NHI Mgmt Group Updated October 31, 2026 Domain: Agentic AI & Autonomous Identity

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.

What Agentic Computing Changes in Security

Agentic computing shifts AI from passive assistance to autonomous action. That matters because the security question is no longer only whether a model can answer correctly, but whether an AI actor can be trusted to decide, initiate, and complete work inside live systems without creating unsafe side effects.

Compared with ordinary software, the security surface expands to include decision rights, tool use, delegation, and runtime policy enforcement. An agent may behave differently from one request to the next, so controls have to be evaluated as dynamic governance rather than static application configuration.

Authentication, Authorization, and Audit for Autonomous Actors

The most important control shift is that agent identity has to be treated as an operating subject, not just a technical detail. An agent should authenticate as a distinct actor, receive scoped authorization for specific tasks, and leave audit evidence that shows what it did, on whose behalf, and under what policy.

That is why identity design for agents often includes delegation boundaries, step-up approval for sensitive actions, and explicit separation between the human principal and the autonomous actor. NHIMG’s Agentic AI Identity Guide explains how those identity choices change as the agent moves from simple assistance to delegated execution.

When organisations need broader governance context, the Agentic AI Security Guide shows how identity, tools, memory, and orchestration fit together in a layered control model. For implementation detail around authentication, delegation, and multi-hop trust, Multi-Agent and A2A Security Guide is the clearest companion resource.

Why Runtime Governance Matters More Than Static Policy

Agentic systems can change plans, chain tools, and take different paths to the same outcome. That means a one-time approval or a fixed role assignment is rarely enough on its own. Practitioners need controls that follow the action, not just the account.

This is the same reason zero standing privilege, continuous evaluation, and per-action policy checks are so central to agentic design. A safe agent is one whose permissions are narrow, time-bound where possible, and revalidated when the action becomes sensitive or context changes.

NHIMG’s Zero Trust for AI Agents is useful here because it frames the problem as continuous verification of the agent, the principal, and the request. The Agentic AI Security Policy Template then turns that idea into governance around registration, oversight, monitoring, and retirement.

Operating Model, Inventory, and Control Ownership

Agentic computing also changes how organisations discover and govern software that acts on their behalf. If agents are created by teams, embedded in workflows, or connected through SaaS and automation platforms, the first failure is often not technical exploitation but lack of inventory, unclear ownership, or shadow deployment.

That is why control ownership needs to be explicit: who can create the agent, who approves its access, who reviews its logs, and who retires it when the business process changes. The most mature programmes treat agent lifecycle management as part of the broader access and assurance model, not as an optional AI add-on.

For discovery and governance, the Shadow AI and AI Agent Discovery Guide helps find unmanaged agents and the grants that empower them. For policy and standards context, the Agent Identity Standards Tracker is useful when teams need to compare the moving landscape of identity and federation options.

Risk and Threat Considerations

Agentic computing increases the risk of overreach because an agent can combine legitimate access with unexpected judgment, chaining, or persistence. The main danger is not that the agent is always malicious, but that a valid actor can be steered into actions that exceed the human’s intent, approved scope, or monitoring assumptions.

Failure mechanism: An attacker, prompt injection path, compromised connector, or mistaken policy can cause the agent to misuse tools, escalate privileges, leak credentials, or execute actions in a workflow the operator did not anticipate.

Impact: The result can be unauthorized access, data exfiltration, destructive changes, fraudulent transactions, or lateral movement across systems with the appearance of legitimate activity, which makes detection and rollback harder.

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 addresses the attack and risk surface, while NIST SP 800-53 Rev 5 and NIST Zero Trust (SP 800-207) set the governance and control requirements practitioners need to meet.

FrameworkControl / ReferenceRelevance
OWASP Agentic AI Top 10ASI03 — Identity & Privilege AbuseAgentic computing centers delegated identity and privilege for autonomous actors.
Recommendation — Enforce per-action authorization and scope agent privileges to the minimum needed.
NIST SP 800-53 Rev 5IA-9 — Identification and Authentication (Service and Device Accounts)Autonomous agents act as non-human software actors that must authenticate distinctly.
AU-2 — Event LoggingAgentic systems need attributable logs for autonomous decisions and actions.
AC-6 — Least PrivilegeAutonomous agents are safest when their access is narrowly scoped and time limited.
Recommendation — Authenticate each agent as a distinct software actor and bind it to auditable credentials. Log agent prompts, decisions, tool calls, and outcomes for attribution and review. Limit each agent to the minimum permissions needed for its current task.
NIST Zero Trust (SP 800-207)PR.AA-05 — Policy enforcement for access decisionsAgent actions should be checked dynamically rather than trusted on standing access alone.
Recommendation — Apply per-request policy checks before an agent can invoke tools or change state.

Practitioner Guidance

Why practitioners should care: Agentic computing is only safe when the organisation can prove what each agent is allowed to do, when that permission applies, and how the action is attributable after the fact. Without that, autonomy becomes an uncontrolled privilege layer rather than a productivity gain.

Governance implication: Assign an owner for every agent, require a named business purpose, and define the approval and review rules for its actions before it is connected to production systems. The key question is not whether the agent is useful, but whether its authority is bounded enough to survive real-world failure.

Practitioner takeaway: Treat agent identity, runtime authorization, and auditability as one control plane, not three separate problems.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Reviewed and updated by the NHIMG editorial team on October 31, 2026.
NHI Mgmt Group — the #1 independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org