AI Portfolio Assistant
Consider a financial institution providing portfolio managers with an AI Portfolio Assistant operating across a suite of FDC3 applications. The assistant can access the user's portfolio, invoke specialist applications to calculate exposures and stress scenarios, retrieve relevant investment research, and explain the results. Rather than asking the AI to perform financial calculations itself, deterministic applications provide portfolio, risk and market information, while the AI combines and interprets those results to help the portfolio manager understand their positions and risks.
As the assistant becomes more capable, however, the consequences of its actions increase. Reading a portfolio or explaining a risk may require relatively little authority, while recommending that the portfolio manager reduce a position is more consequential. Constructing a proposed set of orders goes further still, and actually submitting those orders would give the AI direct authority over the portfolio. Governance therefore needs to control how much authority is delegated to the AI and under what conditions. An organisation might allow the assistant to analyse and explain autonomously, permit recommendations only when specified controls have been satisfied, require human approval for proposed orders, and prohibit autonomous trade execution entirely.
This example follows that progression from observe → analyse → explain → recommend → propose → execute. It demonstrates how Ten-Factor Governance can make the boundaries around AI authority explicit, machine-readable, enforceable and auditable across a real multi-application workflow. The example also shows how those governance mechanisms can form part of an ISO/IEC 42001-aligned AI management system.
Controls are intended to stack by layer:
- CCC Kubernetes on the shared Managed Kubernetes host (both K8s deployments)
- CCC Multi-Agent Reference Architecture capabilities and controls composed into the AI Portfolio Assistant twin
- CCC Object Storage for the Decisions and Evidence Record bucket
- FINOS AI Governance Framework controls in the shared Governance Model (lexicon → policy), hosted by the Artifact Repository twin
- Runtime evaluation and evidence behaviours in the Governance Runtime twin
- The ISO/IEC 42001 AIMS documents wrap that combined set in a management-system envelope
See deployment for the C4 view and twin links, and Gemara component twins for the catalog tree.