What AI-SPM Actually Does
Wiz AI-SPM is built on agentless cloud scanning. It discovers AI resources across an environment, such as models, pipelines, and data stores, and evaluates their configuration against known risk patterns. The output is a posture score and a list of misconfigurations: overly permissive roles, exposed endpoints, unencrypted data paths, and similar findings. This work happens on a scan cycle, not continuously, and it inspects the state of resources rather than their behavior.
Why Configuration Scanning Cannot See Runtime Behavior
A posture scan can confirm that an IAM role is broader than it should be. It cannot see what an agent does with that role during a live session: which tools it calls, what data it retrieves, or whether its behavior deviates from expected use. Because scanning is periodic and agentless, there is an inherent gap between when a risky configuration exists and when it is discovered, and no visibility at all into what happens inside that window during active agent execution.
What Runtime Enforcement Adds
Runtime enforcement operates at the point of tool execution. It evaluates each action an agent identity attempts to take, such as calling a tool, accessing a data source, or invoking another agent, against a live policy, and can allow, deny, or modify that action in the moment. This requires treating the agent as a distinct identity with its own permission scope, rather than as an extension of a human user or a shared service account, so that enforcement decisions and audit logs are attributable to the correct actor.
Posture Management vs Runtime Enforcement
| Capability | AI-SPM (Wiz) | Runtime Enforcement (Trussed) |
|---|---|---|
| Detection method | Periodic, agentless cloud scanning | In-session evaluation of each tool call |
| What it evaluates | Configuration and permissions at rest | Live actions taken by an agent identity |
| Response to risk | Flags and scores after the fact | Allows, denies, or modifies in real time |
| Agent identity model | Not required for scanning | Distinct non-human identity with defined scope |
| Audit evidence | Configuration compliance record | Record of what was actually executed |
Failure Modes Posture Tools Cannot Address
Several categories of AI agent risk manifest only during live execution and leave no trace in cloud configuration state: prompt injection that alters an agent's behavior mid-session, excessive agency where an agent takes actions beyond its intended scope, and credential reuse across agents that obscures which identity actually performed an action. Posture scanning was not designed to observe any of these, since they occur in prompt content and tool-selection behavior rather than in resource configuration.
Where the Two Categories Are Complementary
Posture management and runtime enforcement are not competing approaches; they cover adjacent parts of the same risk surface. Posture findings, such as an overly broad IAM role, can inform runtime policy so that enforcement decisions account for known weaknesses rather than treating every agent identically. Together they provide both a record of what was authorized and a record of what was actually done, which is typically what compliance and incident response require.