E23 ATLAS
A clearer view of model & AI risk.

Discover models and AI usages, maintain an evidence-backed inventory, and
review changes with human judgment at the center.

WHAT ATLAS SUPPORTS
- Read-only technical discovery from source repositories (Azure DevOps, GitHub,
  Gitea), model registries (MLflow, Databricks Unity Catalog), Azure AI services
  and agents, SharePoint and OneDrive, spreadsheets, and ServiceNow, Confluence
  and procurement exports.
- Evidence-bearing candidate triage and owner confirmation.
- Separate governed Model, ModelVersion and ModelUsage records.
- Institution-configured risk methodology with human approval.
- Evidence provenance, lifecycle history and authorized structured search.
- Complete-source reconciliation and reasoned human change dispositions.
- Reviewed imports and exports that complement existing governance tooling,
  and a full tenant export for exit planning (off by default; deletion
  afterwards is an operator procedure).
- One inbox for owner and governance work, with in-app notices by default and
  email through Microsoft Graph or SMTP once an institution approves it.
- Optional modules for third-party risk (B-10), critical operations and
  resilience (E-21 section 3) and technology risk (B-13). They are off by
  default.
- A Canadian French interface, as a machine-assisted draft that a native
  francophone must review before customer use.

A FOCUSED EVALUATION
1. Agree an approved source scope and data handling requirements.
2. Review discovered candidates and their original evidence.
3. Compare the findings with the current governed inventory.
4. Agree follow-up governance actions and evaluation criteria.

PRODUCT PRINCIPLES
AI proposes; authorized people approve consequential governance decisions.
Source connections are read-only within explicitly approved scopes.
Evidence and material decisions retain provenance and audit history.
Atlas supports governance. It does not certify regulatory compliance.
Everything above is verified only with synthetic data and local test servers.

All diagrams on the landing page are illustrative examples, not customer data.
Customer activation is subject to institution-specific policy, security and
operational approvals. This overview does not authorize source access.