Notifications
Clear all
Topic starter
06/09/2026 3:45 pm
TL;DR: AI agents turn model interaction into system action, so identity, permissions, tool access, and auditability become the centre of AI-SPM, according to LEVO. The practical shift is from static permission thinking to runtime control over tool chains, approvals, and evidence, because traditional workflow assumptions do not fully contain agent behaviour.
NHIMG editorial — based on content published by LEVO: LLMjacking: How Attackers Hijack AI Using Compromised NHIs
By the numbers:
- 80% of identity breaches involved compromised non-human identities such as service accounts and API keys.
- 97% of NHIs carry excessive privileges, increasing unauthorised access and broadening the attack surface.
- Only 5.7% of organisations have full visibility into their service accounts.
Questions worth separating out
Q: What breaks when AI agents are given access without identity governance?
A: What breaks is accountability.
Q: Why do AI agents increase risk when they can move from read to write actions in one workflow?
A: Because static permission models assume the risk boundary is known in advance, but an agent can chain calls and reach higher-impact tools at runtime.
Q: What are the signs that agent posture controls are failing?
A: Look for unusual write volume, spikes in tool calls, repeated failures, and actions that occur without a clear approval trail.
Practitioner guidance
- Separate agent, user, and tool identities Create distinct identities for the human requester, the agent runtime, and each downstream tool or service account.
- Build a tool permission matrix Classify every tool as read-only, low-impact write, or high-impact write, then require explicit allowlists for the second and third tiers.
- Enforce transaction boundaries for multi-step actions Add confirmation gates, timeout budgets, and loop detection before the agent can repeat actions or cross into irreversible operations such as deploys, deletes, or payments.
What's in the full article
LEVO's full blog post covers the operational detail this post intentionally leaves for the source:
- A practical agent posture model with identity, permission, tool-chain, data, execution, and monitoring layers
- A detailed tool permission matrix that separates read, write, and admin actions across risk tiers
- Examples of rate limits, timeout budgets, loop detection, and confirmation gates for irreversible actions
- A testing approach that maps prompt injection and output risks to OWASP LLM Top 10 and MITRE ATLAS
👉 Read LEVO's analysis of agent posture management and AI-SPM controls →
Agent posture management: are your AI controls keeping up?
Explore further
06/09/2026 3:45 pm
Agent posture management is really delegated-identity governance for AI agents. The article is describing a control model built around identity, permissions, tool catalogs, approvals, and audit trails, which is the same governance problem NHI teams already face with service accounts and API keys. The difference is that agents make the runtime path more dynamic, so the actor chain becomes a first-class security object. Practitioners should treat this as an identity programme problem, not a prompt-layer problem.
A few things that frame the scale:
- Only 5.7% of organisations have full visibility into their service accounts, according to the Ultimate Guide to NHIs.
- 79% of organisations have experienced secrets leaks, with 77% of these incidents resulting in tangible damage.
A question worth separating out:
Q: How should security teams govern AI agents that call internal tools from Kubernetes?
A: They should separate runtime containment from identity governance. Kubernetes can isolate execution, but the tool boundary still needs delegated identity, per-request authorization, and action-level logging. Without those controls, the agent behaves like an unscoped service account with model-driven output, which makes attribution and revocation much harder.
👉 Read our full editorial: Agent posture management reframes AI security around identity and actions
