Runtime Action Type Matrix: Which Tool Stops Which Action
The Runtime Action Type Matrix lists 16 AI runtime security tools against 8 harmful actions: shell command, file access, data paste, package install, network call, credential access, prompt injection and harmful output. Each cell is yes, partial or none found, from vendor evidence read on 16 Nov 2026.
Facts read on .
The matrix
- Yes: documented specifically
- Partial: documented in general terms, for some agents or clients only, or detect and alert without blocking
- None found: none found on 16 Nov 2026
What does the matrix show?
The matrix shows, for each of 16 vendors, whether the vendor documents stopping each of 8 action types before the action completes. A yes needs a specific description in the vendor's own pages or a primary report; partial covers general wording, limited agent coverage, or detect-and-alert without blocking.
Use it to match a tool to the action that worries you most. A tool that stops shell commands on a laptop may say nothing about a model reply, and the reverse holds for filters at the model layer.
How to use the action picker
The home page action picker reads this matrix. Choose an action and see which tools document stopping it. Prompt injection and harmful output each have their own page with the reasons and the layers where they can be stopped; the six machine actions are columns of this matrix and rows of the picker.
Download and cite
Download the Runtime Action Type Matrix as CSV or JSON. Cite it as: AI Runtime Security Tools, Runtime Action Type Matrix and Enforcement Point Map, edition 16 Nov 2026. https://airuntimesecuritytools.com/. Licensed CC BY 4.0.