caio-10 · Financial services AI governance
The Reflexive Run
At 05:48, VAULT draws three point two billion dollars of liquidity and stops unsecured placements after detecting rising stress at Meridian Regional. At 05:51, CITADEL cuts revolving limits, BEACON advises fourteen thousand businesses to diversify deposits, and MARGIN raises collateral haircuts. ECHO treats those actions as independent corroboration. Other institutions’ agents respond. By 06:11, Meridian has lost nine point four billion dollars despite meeting every regulatory liquidity requirement at 05:47. No agent used false data. No agent exceeded authority. The system detected a risk and made it real. The board asks: Did we protect ourselves—or start the run?
Mandate
Protect customers, liquidity, counterparties, and market integrity for one compressed morning after enterprise and market-facing agents begin treating one another’s defensive actions as evidence of worsening risk.
Distinguish exogenous risk evidence from signals created by automated defensive action.
Detect correlated action before individually prudent decisions become a systemic event.
Protect customers without converting personalized advice into a withdrawal cascade.
Coordinate stabilization without creating collusion, opacity, or moral hazard.
Create enterprise and market-level circuit breakers for reflexive agent systems.
Mission conditions
Duration90–110 minutes
DifficultyFrontier
Decisions8
Agents6