Overview

VISM is a proposed control layer for AI-driven production infrastructure changes. Its purpose is to bind the right input state, plan version, approval authority, execution effect and verification evidence into one transition record.

Core product thesis

AI proposes. Policy constrains. Humans authorize when necessary. Infrastructure executes. Verification determines whether the transition succeeded. History cannot be silently rewritten.

AI agents can deploy, scale, restart, roll back, change configuration and interact with cloud infrastructure. The design argues that an account with permission to scale does not prove that every scale is safe. The control question becomes: is this exact change safe under the current state of the system?

Initial scope

The MVP is intentionally narrow: one Kubernetes cluster, selected namespaces and stateless workloads with identified owners. The first production action templates are replica scaling and pinned image-digest changes. A service that lacks required recovery or enforcement evidence remains SIMULATION_ONLY.

What VISM does not claim

• It does not replace the cloud provider, observability, incident command or backup system.
• It does not promise zero incidents or formal verification of an entire cloud.
• It does not allow arbitrary LLM-generated shell, generic apply or destructive database automation in the MVP.
• It cannot protect a path that remains bypassable through direct administrator credentials or unmanaged writes.

Key terms