Recon handles sensitive financial documents. Here is how the platform is designed to keep that data protected, keep humans accountable, and keep a record of every decision.
Agents produce evidence, not decisions. No AI in Recon can approve or reject anything — the binding call is always a human action, recorded against the case.
Recon runs as independent regional silos. A customer’s data lives in exactly one region — Europe / West Africa (Paris) or Asia-Pacific (Sydney) — and never crosses to the other. Each region is a fully separate stack: its own database, storage, and secrets. Your organisation is pinned to its region at onboarding.
The reconciliation and document-validation agents run deterministic checks and surface findings. The decision to approve, reject, or escalate a case is a person’s — and it is written to a durable audit record with who decided, what, and when. This is a deliberate posture for responsible-lending and fair-processing obligations.
Security-relevant actions — case decisions, administrator changes, account erasure — are written to an append-only audit log, separate from application logs. Structured, machine-readable logging underpins monitoring and investigation.
Recon supports self-service and administrator-mediated account deletion that removes a user’s documents from storage and cascades their records — for GDPR and Privacy-Act right-to-erasure requests. A member’s erasure never touches a colleague’s documents or shared reference material.
Every data read is scoped to your organisation at the database layer — enforced in code, not left to the model. Sign-in is passwordless magic-link, with an administrator-managed access allowlist and role-gated admin functions.
Recon is in active development. This page describes the platform’s security architecture and posture; formal certifications and a detailed data-processing agreement are available on request as we work with institutional customers.