Research

Our work spans model integrity, adversarial threats, machine trust and security evidence. These are workstreams, not products.

Four research areas

Area	What it studies	With Vraelis Contour
Model integrity	Whether the model that runs is the model that was approved.	Draws on directly. Release identity, bound approval, approved against reported.
Adversarial threats	Inputs, updates and supply paths built to change what a model does.	Informs. Which supply paths a release check covers.
Machine trust	Explicit identity, limited permissions and checks at execution boundaries.	Informs. Separate approval roles, a policy checked where it serves.
Security evidence	Records a person can read, kept in order, including what could not be confirmed.	Draws on directly. Ordered observations, including unavailable evidence.

Vraelis Contour is the first product direction, and it draws on model integrity and security evidence directly. The research areas are not products and have no product claims of their own.

Guides

Three guides describe how we think about model releases, operating constraints and AI security evaluation. They are research guidance, not product capability.

Model release securityFor the engineers who move an approved model into an operational system.Read Operating constraintsThe control has to fit the environment, not just the model.Read AI security evaluationMatch the threat to a measurable test.Read

Last updated: October 10, 2026