Independent analysis by NHI Mgmt GroupSource article: Teleport, “Introducing Teleport Beams: A Trusted Runtime for Securing Agents in Production Infrastructure” (November 10, 2026)

TL;DR: Teleport says Beams gives AI agents isolated runtimes, delegated identities, short-lived access and full audit trails so production work can run without sharing human credentials or leaving stale secrets behind. Access review and privilege controls have to move from user-centred governance to task-scoped runtime enforcement when agents act independently.


At a glance

What this is: NHI Mgmt Group's analysis, drawing on Teleport's blog post: AI agent production access needs isolated runtimes, delegated identity and short-lived privilege to avoid shared-credential risk.

Why it matters: Identity teams now have to govern agent execution like a live privilege boundary, not a generic automation layer, because runtime decisions and access scope can change mid-task.

👉 Read Teleport's analysis of production identity controls for AI agents


Context

AI agents that can act inside production systems create a governance problem that standard automation controls do not solve: the actor can decide, request tools and use access in the same session. That breaks assumptions built around stable users, durable service accounts and human approval loops. Zero trust for AI agents is the right baseline for understanding why task-scoped identity matters here.

The core issue is not simply whether an agent is allowed to run. It is whether access can be bounded to the task, the session and the specific resource path without inheriting a broad human role. Once the agent becomes a runtime actor in infrastructure, identity governance has to follow the action, not just the account.


Key questions

Q: How do agents get delegated identity instead of shared credentials?

A: Agents should be issued distinct machine-bound identities that map to a specific delegation event, not a reused human login or a broad service account. That model gives the organisation a traceable principal for the agent, keeps access scope tied to a task, and makes it possible to audit who or what authorised the action.

Q: When does just-in-time access help most for AI agents?

A: JIT access helps most when agent tasks are episodic, high-risk, or difficult to predict in advance. It reduces standing exposure, but it works only if the agent's permissions are also tightly scoped and actively revoked after use. Otherwise, the temporary token masks a persistent privilege problem.

Q: How should security teams log AI agent actions for audit and compliance?

A: Security teams should log AI agent actions as identity events, not just application events. Each record should include the human initiator, agent identity, approved session scope, tool invocation details, and any downstream delegation. That structure lets investigators prove whether the action stayed within authorised boundaries and gives compliance teams a defensible record of accountability.

Q: What is the difference between sandboxing and runtime governance for agents?

A: Sandboxing confines execution, but runtime governance decides whether the agent may act on a discovered path in the first place. A sandbox can limit where code runs, yet still leave an agent free to reason about tools and permissions. Runtime governance is the control that evaluates intent and stops unsafe actions before they reach sensitive systems.


How it works in practice

Agent runtime identity versus inherited credentials

An AI agent that inherits a human account or a broad workload role blurs accountability and expands blast radius. A delegated identity model gives the agent a distinct credential binding, so its actions can be traced back to the delegation event rather than to a shared principal. In practice, this separates authentication from authorisation scope and makes the agent a governed subject rather than a disguised proxy for a person or service account. Read Zero Trust for AI Agents.

Practical implication: Treat agent identity as its own governed principal and stop letting agents operate under shared human credentials.

Why short-lived access matters more for agents

Just-in-time access works for agents only when elevation is tied to a narrowly defined task and expires automatically after the work completes. That matters because agent runtime is dynamic: tool choice, timing and action sequence can all shift during execution. Short-lived access reduces the window in which a misaligned or compromised agent can continue to act beyond the intended scope.

Practical implication: Enforce time-boxed elevation for agent actions and expire access when the task ends, not when a manual review happens later.

Observability for prompts, tools and protocol traffic

Agent governance needs more than authentication logs. Effective monitoring has to capture prompts, tool calls, protocol-level traffic and the decision path that led from intent to action. Without that evidence, teams cannot distinguish expected automation from data movement, privilege drift or self-triggered escalation. The runtime becomes governable only when the control plane records what the agent asked, what it accessed and what it tried to do. See the AI Agent Observability, Audit and Incident Response Guide.

Practical implication: Log agent prompts, tool usage and resource access together so review can reconstruct each action chain.


NHI Mgmt Group analysis

Shared human credentials are the wrong trust model for agentic production work: When an agent inherits a user or service account, auditability collapses into the behaviour of a principal that was never designed to represent machine intent. That makes least privilege impossible to enforce at the task level and turns agent activity into an opaque extension of someone else’s access. Practitioners should treat delegated identity as the minimum viable unit of accountability for agents.

Task-scoped access is the control plane, not an optimisation: The meaningful decision is not whether an agent can work faster than a human. It is whether the access it uses can be constrained to one bounded purpose, one resource path and one expiry condition. That is why runtime enforcement matters more than policy statements about responsible use; the control must operate where the action happens.

Standing privilege assumptions fail when the actor can request elevation during execution: Standing privilege is designed for identities whose access is established before work begins and then reviewed later. That assumption fails when the actor can ask for access, receive it and act inside a single workflow. The implication is that governance has to move from periodic review to per-action decisioning.

Agent observability is now an identity requirement: Prompt logs, tool traces and protocol capture are no longer optional telemetry for advanced teams. They are the evidence trail that lets organisations attribute action, detect scope drift and prove whether an agent stayed within its delegated purpose. Without that trail, agent security becomes a claim rather than a control.

Identity governance for agents now intersects with zero trust architecture: The relevant question is whether every agent action is independently verified, authorised and bounded instead of inheriting ambient trust from the surrounding environment. That aligns the problem with zero trust for AI agents rather than with conventional automation governance. The practical conclusion is to govern each request as a fresh trust event.

From our research library:

What this signals

Identity governance for agents is shifting from account-centric review to action-centric control: The old model assumes a principal exists long enough to be reviewed after the fact. With agents, the decision point moves into the runtime, where access can be requested, used and discarded inside a single workflow. Organisations that keep relying on periodic certification will miss the control point entirely.

A useful way to think about this is as an identity blast radius problem: the real risk is not only what an agent can reach, but how quickly it can combine access, tool use and protocol interaction before any human review can intervene. According to the 2026 Infrastructure Identity Survey, only 44% of organisations have implemented any policies to manage their AI agents, despite 92% agreeing that governing AI agents is critical to enterprise security. That gap means governance has not yet caught up with runtime behaviour.

Task-scoped trust debt: A delegated identity may be technically short-lived, but if organisations do not record prompts, tool calls and access paths, they still accumulate unresolved trust debt across each agent session. The practical response is to build evidence collection into the runtime itself, so oversight follows the action instead of arriving after the fact.


For practitioners

  • Define agent identities as separate principals Issue each agent its own delegated identity and remove reliance on shared human credentials or broad workload roles.
  • Scope elevation to a single task Require short-lived, task-specific access requests for agent work and expire them automatically when the task completes.
  • Capture full agent action evidence Log prompts, tool calls, protocol traffic and resource access in one record so investigators can reconstruct intent and outcome.
  • Review production exposure paths Inventory which cloud accounts, databases and tool endpoints an agent can reach, then remove any path that is broader than the task requires.

Key takeaways

  • AI agents in production need their own governed identity and access boundary, not borrowed human credentials.
  • The central risk is not autonomy by itself but runtime access that can widen, change or outlive the task.
  • Auditability depends on capturing prompts, tool use and resource access together, because that is where agent accountability lives.

Related NHI Mgmt Group resources: Zero Trust for AI Agents · AI Agent Observability, Audit and Incident Response Guide · Top 10 Agentic AI Identity Issues · OWASP Agentic Applications Top 10


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 CSF 2.0 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 AbuseThe article centers on agent identity, delegated privilege and runtime access scope.
Recommendation — Apply ASI03 to verify agent identity, constrain delegated privilege and block inherited ambient access.
OWASP Non-Human Identity Top 10NHI-05 — Overprivileged NHIThe core risk is agents running with broader access than their task requires.
NHI-07 — Long-Lived SecretsThe runtime design aims to avoid stale credentials and long-lived access paths for agents.
Recommendation — Scope agent identities to the minimum task privilege and revoke any ambient access. Replace persistent agent credentials with short-lived access and automatic expiry.
NIST CSF 2.0PR.AA-05 — Access Permissions, Entitlements and AuthorizationsThe post is fundamentally about controlling agent permissions and entitlements in production.
Recommendation — Review agent entitlements against task scope and remove any standing permission that is not required.
NIST Zero Trust (SP 800-207)Section 4 — Zero Trust PrinciplesThe runtime applies per-request trust and removes implicit confidence in the surrounding environment.
Recommendation — Enforce per-action verification and assume breach for every agent request.

Key terms

  • AI Agents: AI agents are autonomous software entities that act within organisational environments and make runtime decisions within assigned boundaries. They can hold identities, authenticate to systems, and exercise permissions, which makes them comparable to other non-human identities that require inventory, governance, and continuous activity monitoring.
  • Delegated Identity: Delegated identity is when one actor acts on behalf of another with explicit permission and bounded authority. In AI-assisted commerce, it requires clear consent, limited scope, and traceable records so the retailer can distinguish authorised delegation from unauthorised automation.
  • Just-in-Time (JIT) Access: Just-in-Time (JIT) Access grants permissions only when needed and removes them after use. In IAM and PAM, it reduces standing privilege by issuing time-bound access based on approval, context, or policy, limiting exposure for humans, service accounts, and non-human identities while improving auditability and control.
  • MicroVM: A microVM is a lightweight virtual machine that provides hardware-backed isolation with a much smaller footprint than a traditional VM. In this context it gives an AI agent its own guest kernel, which is a stronger containment boundary than a container because host kernel compromise is no longer the default failure mode.

What's in the full announcement

Teleport's full blog post covers the operational detail this post intentionally leaves for the source:

  • How Beams binds delegated identity to each agent and avoids shared human credentials
  • The runtime controls behind micro-VM isolation, protocol capture and auto-expiring access
  • Examples of agent-initiated JIT requests and how reviewers approve them
  • How MCP access is constrained with short-lived certificates and tool-level RBAC

👉 Teleport's full post covers delegated identity, JIT elevation, runtime isolation and MCP access controls for agents.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an identity security programme, it is worth exploring.
NHIMG Editorial Note
Published by the NHIMG editorial team on November 16, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org