home / threat-models

Threat models

Each finding is published with the model it was found under: what is trusted, what is not, and where the boundary sits.

Models I work under

published alongside each finding

Code inside the sandbox, container or VM is fully attacker-controlled. What can reach the host?

boundary: guest → host

A dependency or build input is malicious. Can the artifact be independently rebuilt and verified?

boundary: source → artifact

Access to the device, firmware or removable media. Do boot integrity and the air gap still hold?

boundary: media → machine

Web requests, tool outputs and model inputs are attacker-influenced. What privileged action do they reach?

boundary: input → action