Trust

Governing how AI is used in reporting.

AI is used in Datox for interpretation and investigation, reading documents, proposing mappings, explaining exceptions. It is not used to invent figures, override rules or approve outputs.

Boundaries

What AI does and does not do

The boundary is drawn around determinism. Where a result must be identical every time and provable, it is computed by configured logic. Where the task is interpretation, an agent does the work and its output is verified.

Explicit limits

AI does not compute reported figures where deterministic logic applies.
AI does not change reporting rules or thresholds.
AI does not clear its own exceptions.
AI does not approve or submit an output.
AI proposals affecting reporting logic require human acceptance.

Assurance

How agent output is controlled

Evidence grounding

Extracted values are tied to their location in the source document or record so a reviewer can confirm them.

Confidence handling

Low-confidence output is routed to review rather than passed through silently.

Deterministic checks

Agent output passes the same validation rules as any other data before it can be used.

Audit record

Agent actions are logged like any other step, including what was proposed and who accepted it.

Data

How your data is used

Client data is processed to produce your reporting. Specific terms covering data usage, model providers, retention and confidentiality are set out contractually.

Client data is processed for the purpose of operating your reporting workflows.
Model and processing arrangements are documented for governance review.
Data handling commitments are contractual rather than implied by this page.

Datox provides software. Responsibility for the accuracy and submission of reported information remains with the reporting entity.