Skip to content
Arbitex Docs

Model Risk Management

What it does

Section titled “What it does”

Model Risk Management is the model-governance workbench: it groups your Model Catalog by family, lets you set a risk tier and validation status at the family or individual-version level, and run an approve / reject workflow with an auditable revision history. A rejected or deprecated model is blocked at dispatch, fail-closed.

Configure in the UI

Section titled “Configure in the UI”
The Model Risk Management panel in the admin console, showing the model inventory grouped by family with Risk tier, Validation, and Approval columns, and Approve/Reject buttons.
Models & Routing → Model Risk ManagementCitadel synthetic data on a non-production Arbitex demo instance. Never client data.

  1. Open Models & Routing → Model Risk Management. Overview cards show totals for Families, Models, Approved, High risk, and Critical risk, alongside charts of models by risk tier and by provider.

  2. In the inventory table, set Risk tier — Unassigned, Low, Medium, High, or Critical — and Validation — Unvalidated, Pending, Validated, or Failed — at the family level. This cascades to every version in that family; expand a family (▸) to override a single version instead. A family with mixed values across its versions shows a Mixed badge.

  3. Select Approve or Reject on a family or version row and confirm in the dialog. Both actions commit an immutable, audited revision — there is no separate justification field, just the confirmation.

  4. Review the Approval history grid at the bottom: each row is a revision, when it was committed, and how many records it covered — the full auditable trail of governance changes.

API equivalent

Section titled “API equivalent”
Terminal window
curl -X POST "https://admin.arbitex.ai/api/v1/admin/models-routing/model-risk/list" \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"du_id": null}'

Approve and reject are both expressed as the same authoring call — there is no separate approve/reject endpoint:

Terminal window
curl -X POST "https://admin.arbitex.ai/api/v1/admin/models-routing/model-risk/set" \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"du_id": null, "objects": [{"object_key": "claude-3-5-sonnet", "risk_tier": "medium", "validation_status": "validated", "approval_status": "approved"}]}'
Terminal window
curl -X POST "https://admin.arbitex.ai/api/v1/admin/models-routing/model-risk/revisions" \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"du_id": null}'
# → the approval-history revision log

Plan & availability

Section titled “Plan & availability”

No plan-tier gating is visible in the console for Model Risk Management itself — it is available to every admin regardless of plan. Broader compliance workflows (SOC 2 / HIPAA reporting) that this governance data feeds into are gated per your plan tier; see Billing & Plans.

Deployment notes

Section titled “Deployment notes”

Risk tiers, validation status, and approval decisions are authored once and apply to every deployment unit serving your organization, including a Hybrid Outpost — a rejected model is blocked at dispatch at every outpost, not just the one where you approved or rejected it.

Section titled “Related”