Governance before computation
Human oversight, traceability and sovereignty are not features added after the model. They shape the processing pipeline from the first input.
Intelligence must never justify a data leak
Processing location depends on confidentiality and re-identification risk, not only on model power.
Real-world privacy and compliance must still be evaluated before any claim of certification.
Private inference only. Personal and Proreta financial records are never sent to external neural services.
External compute may be permitted under explicit rules, with logging and flow control.
Verified filtering and anonymization; deny by default where re-identification remains possible.
Human experts retain responsibility. Machine summaries are not validated evidence.
Power without control is not progress.
Auditoris separates what must remain separate, connects what may be connected and makes limitations explicit.