Skip to content
Critical infrastructure

AI security for critical infrastructure.

In energy, water and communications, vendors ship models into systems operators own.

Electrical equipment at a power substation.

Intended application areas. Vraelis has no customer deployments in them.

Why it matters

The vendor ships it. The operator runs it.

Releases cross operators

Both sides need to know which release is running.

An unapproved release is a security event

A change to an operational system, to be handled as one.

The change record needs the model

What was approved, for where, and what the runtime reported.

How Vraelis Contour would apply

Five steps, release to record.

How Vraelis Contour works
  1. Identify the release

    Working in the experiment: model and configuration digestsProposed: preprocessing identity
  2. Bind approval to the release and destination

    Destination-bound approval: one site's approval does not cover another.

    Working in the experiment: signed bindings
  3. Integrate with a supported runtime

    Working in the experiment: ONNX Runtime 1.23.2, CPUProposed: customer runtime adapters
  4. Compare approved and reported

    Working in the experiment: evidence app
  5. Preserve discrepancies, missing evidence and recovery

    Deployment evidence kept for the operator's change process.

    Working in the experiment: recorded observationsProposed: investigation workflow
No loaded report is available for this observation.
Read-only evidence app, inside the controlled ONNX reference experiment.
Limits

What we do not claim.

  • No deployments with utilities or other operators, and no customers.
  • No integration with control systems or operational technology networks.
  • No replacement for an operator's change management or protective controls.
  • No certification or regulatory compliance.
  • No hardware attestation. Runtime identity is a worker self-report.

Build stronger evidence for model releases.

Join us in building Vraelis Contour.