AI governance

The service never makes or approves a decision. It produces argument, tension and a structured report for a human to interrogate. Every report is framed as contested analysis, including where the evidence is thin.

Personas are composite constructs. They are not digital twins of named individuals and are not presented as statements of any real person's views. Where a persona is derived from a real participant with their involvement, that origin is recorded and access is restricted.

Persona behaviour derives from recorded identity, experience and knowledge records rather than free invention, and debates are grounded in the scenario and evidence supplied. Gaps in evidence are surfaced to the user for review before a debate is run.

Changes to canonical persona records pass through an administrative review queue, and edits, learning events and deletions are logged. Users see their own history; administrators see the workspace history.

The service tracks whether personas discriminate meaningfully rather than converging on a single voice, and flags profiles that are too shallow to be trusted. Repetition and template drift are monitored and corrected.

The service must not be used to make automated decisions about individuals, to profile a named person for adverse action, to simulate a specific identifiable individual without their consent, or to generate material presented as that person's real statements.

Concerns about output, persona construction or misuse can be raised through the enquiry form. Escalations are reviewed by the operator and can result in persona withdrawal or account suspension.