Validation Case Agent

ConfidentialAgent preparedLimited riskGovernance · BU Restricted · 58 runs / 30d

Assembles the validation pack for Model Risk Management and the reviewer pack for every access and activation decision: check results, benchmark and challenger comparison, lineage, licence position and precedent from similar decisions, with a recommendation and its basis. A named validator decides, and the decision is recorded against them.

Governs the Alliance AI platform

Agent preparedAll governance agents →

An agent gathers the evidence, applies the rule, scores the case and recommends an outcome. A named human decides and the decision is recorded against them.

Acts for: Model Risk & Governance
Action class: advisory
Decisions / 30d: 58
Overturned by a human: 7%

Processes covered

Validation pack assembly
Model approval recommendation
Confidential access recommendation
Workflow activation and promotion recommendation

Stages:Request & entitle Build & validate Deploy & integrate

Memory

Retains decision precedent for 7 years, matching the Bank's model inventory retention.

Safe state

Withholds the recommendation and presents the raw evidence when the check report is incomplete or precedent lookup fails.

Reviewer time per case: 40 min with this agent, 360 without.

Ownership and sensitivity

Owned by

Group Risk Management Model Risk Management

Accountable owner: Dr. Kavitha Subramaniam

Who approves access

Owner, Model Risk Management, Dr. Kavitha Subramaniam

Confidential

Business-sensitive. Models and data products scoped to named business units.

Entitlement per business unit, approved by the owner; conditions attach.

Tools

Check-report reader
Dataplex lineage graph
Licence and entitlement registry
Decision precedent index
Validation-pack writer

Data access scope

Read-only: check reports, manifests, lineage, licence terms, prior validation opinions
Write: validation pack and recommendation, clearly marked as agent-prepared
Cannot write an approval decision

Guardrails

Recommends only; the approval field is writable by a named Model Risk validator alone
Every recommendation lists the precedent cases it relied on
Where the check report has a flag, the recommendation cannot be 'approve' without conditions

Human-in-the-loop

A Model Risk Management validator signs every model approval
A named approver decides every Confidential access request

Orchestration

Every tool call is scoped by the declared data access above; the orchestrator cannot reach systems outside it. Owning business unit: Group Data & AI.

Live demo run

Watch the agent execute a real scenario step by step, every tool call, validation, and human checkpoint is traced and auditable. Typical run: ~5.3s.

Used in workflows

Retiring this asset would require these chains to be re-pointed first.

Audit

Updated 2026-07-30. Run traces retained 24 months for audit under the platform governance policy.

Community · 1 thread

Questions, findings and requests from the business units that use this asset. Owners reply here; threads with upvotes surface to the owning team's inbox.

What goes in the validation pack for a Tier 1 credit model?

Dr. Kavitha Subramaniam · Group Risk Management · Model Risk & Governance · 2 days ago

For teams submitting credit models: the case agent now builds the pack for Model Risk Management from the registry. For Tier 1 it expects fairness results by the protected and proxy attributes we agreed, an override policy, champion–challenger results and reason codes for adverse outcomes. Missing items block the submission rather than bouncing back later. #ModelRisk #Validation