Security & Trust

Control is part of how the system runs.

Enterprises adopt AI when they can control it. Roux treats permissions, approvals, oversight and evidence as runtime behaviour, not documentation.

Control planes

Six areas, designed together.

Each one is enforced where the operation runs, so a control cannot be true on paper and false in production.

Roles, scopes and least-privilege access to operations, data and actions.

Encryption in transit and at rest, with data handled inside defined boundaries.

Every decision, action and approval recorded so it can be reconstructed later.

What the AI may do, which actions need approval and how changes are released.

Supervision, intervention and escalation are part of the runtime, not an add-on.

Isolated environments, controlled deployments and monitored operations.

Evidence

Every operation leaves a record that can be reconstructed later — for a customer dispute, an internal review or an auditor.

What the record contains

Autonomy is granted, never assumed.

Human oversight

Supervision: Operations can be watched while they run, not only after they finish.
Approval: Sensitive actions wait for a named person before they execute.
Intervention: A person can take over an operation mid-flight.
Escalation: Handovers carry the full context, so nothing is re-asked.

Our position

An AI system that cannot be explained cannot be trusted with customers.

We would rather constrain an operation than allow an action nobody can account for.

Access, data handling, oversight and evidence — reviewed against the operation you intend to run.