Running capital with AI demands a culture that is more conservative, not less. A sophisticated model does not remove the need for disciplined controls; it increases the importance of understanding when and why a system may fail.
We approach material model changes like aircraft maintenance: documented, tested, monitored, and reversible.
Operating principles
Decision trails should be transparent and auditable. Limits on leverage, concentration, and model risk should be explicit. Systems should fail safely, and unusual behavior should trigger review rather than silent adaptation.