Security & AI

Your controls.
Your decisions.

Set the scope for optional AI assistance. Keep source figures, reporting responsibilities, and professional judgment within a defined control model.

Organisation controls

Set the scope.
Keep it bounded.

AI availability and provider configuration are organisation decisions. Each cycle and action operates within that approved scope.

01 / Organisation

Set the permitted scope.

AI availability, approved providers, and managed credential references belong to the organisation.

02 / Reporting cycle

Choose whether assistance is used.

A separate cycle setting can switch AI actions off. It cannot expand the organisation’s approved scope.

03 / Named user

Keep the decision with a person.

AI suggestions need a user decision. AI cannot calculate figures, approve work, or advance a review state.

Before an AI request

A defined task.
A checked request.

Approved mapping, statement-review, and wording tasks pass through a bounded gateway. Extra tools, messages, or endpoint changes are rejected.

Inspect the request boundary.

The documented policy masks supported figure formats. A bounded gateway validates the allowed task before transport.

The masking boundary

The documented english-figures-v1 policy covers supported numeric and written-number formats. It does not prove that arbitrary prose is free of financial meaning. Other languages and formats need a separately tested policy.

Optional assistance

AI is a choice.
Review is the workflow.

Keep AI review actions separate from the core reporting process. Confirm the deployment and provider arrangement for your environment.

AI actions enabled

Assistance within scope.

Use defined mapping suggestions, statement review, and wording assistance. The organisation controls provider configuration and credentials.

Suggestions require a named user decision.
AI actions disabled

People lead the review.

Use the reporting workflow without requesting AI review actions. Source tracing, preparation, and checking remain the focus.

AI review requests are switched off.

Your production environment

Confirm the arrangement.

Production AI credentials require an organisation-scoped managed secret reference. Development credential fallbacks are not accepted in production.

Deployment
Hosting model, network boundaries, and operational responsibilities.
Data handling
Retention, encryption, backups, and deletion arrangements.
AI providers
Approved provider, credentials, subprocessors, and provider policies.
Assurance
Security documentation and any certifications evidenced for the deployed environment.

Access & evidence

Follow the action.
Inspect the record.

Authentication, organisation and team access, reporting roles, and server-side authorisation define the working boundary. Artifact versions and audit events retain what the application saw and what people decided.

Explore review controls
What the record establishes

Requests, attempts, responses, validation outcomes, and human decisions. A transmission attempt does not prove provider receipt. Database-only evidence is not tamper-proof against database administrators.

Your governance requirements

Your questions.
A focused walkthrough.

Review the control model, deployment requirements, and AI boundaries with your reporting and technology teams.