Different systems. Different security boundaries.
The control has to fit the environment, not just the model.
This guide describes how we think about the problem. It is not a description of a product capability. Where it mentions Vraelis Contour, it says what exists in the experiment and what is proposed.
Disconnected operation
Mission systems can lose connectivity. Identity, release approvals and evidence need a defined behaviour when central services are unavailable.
Define outages, revocation, operating limits and independent controllers with the system owner.
Established operational controls
AI is introduced alongside existing equipment, permissions and change management processes. Security needs to respect those boundaries.
A policy engine can return a decision. The component holding access to the resource must enforce it against the exact request, the current authority and any required approval.
Limited edge resources
Model updates and security checks must fit the system's compute budget, timing requirements and recovery process.
Edge compute budgets, disconnected operation and supplier interfaces shape the integration. Hardware support, detection performance and deployment suitability must be demonstrated for each environment.
What this means for Vraelis Contour
The experiment runs on a CPU in a lab environment that does not actuate anything. It does not yet handle disconnected operation, report freshness or constrained edge hardware, and those are among the items not built. Limits and what is not built
Read the underlying guidance
Principles for the secure integration of artificial intelligence in operational technology
Last updated: October 10, 2026