Day After Recovery Simulator
Model an agentic incident across the four decision horizons from CABIER White Paper 05: prevent unauthorized authority, detect and contain harmful execution, stabilize services and liquidity, and restore trusted operations with evidence.
CABIER research · LUMINAIRE simulation
Day After Recovery Simulator
The institution
The incident
The four decision horizons
Each horizon is graded from your scenario inputs. Hover a card to inspect its evidence standard.
01
PREVENT
Bound delegated authority
Agent register linked to critical-service maps, hard caps enforced outside the model
Holds in this scenario
02
DETECT
See harmful execution early
Replayable incident trace, alert thresholds, measured time to contain
Partially covered
03
STABILIZE
Protect services and liquidity
Tested restricted operating modes, reconciled balances, confirmed intraday funding
Partially covered
04
RESTORE
Return only with evidence
Independent records, credential rotation, staged restoration with independent approval
Fails in this scenario
Unresolved window
12 h
Detection plus containment time during which harmful execution can continue.
Parallel failures
3 firms
Firms on the shared stack fail together, not one after another.
Confidence shock index
36 / 100
Illustrative index of uncertainty exceeding actual loss. Consistent communication lowers it.
Liquidity shortfall
5%
Expected outflow beyond the usable intraday buffer.
From one compromised agent to a sector event
Simulation 05 from the paper, pre-authored and deterministic.
01
Initiating event
A compromised agent, poisoned instruction source, stolen machine identity or unsafe model update invokes a legitimate tool in an unauthorized way.
02
Parallel failure
Firms sharing the same provider, model stack, connector or identity service fail in parallel rather than through lateral movement.
03
Financial transmission
Delayed, duplicated or unverified transactions. Treasury cannot distinguish actual outflows from queue state. Depositors move; counterparties raise intraday buffers.
04
Confidence shock
Inconsistent communications and synthetic messages push uncertainty beyond the actual loss.
05
Controlled Day After
Restricted operating state, verified record of obligations, staged restoration with independent approval.
The Day After operating sequence
First hours
Command and containment
- •Establish incident command and identify affected agents and identities
- •Impose transaction and privilege limits; isolate compromised tools without disabling unrelated essential services
- •Use pre-agreed contact paths that do not rely on the affected identity or communications environment
- •Reconcile queues, balances, settlement positions, collateral and intraday funding needs from independent records
- •Give supervisors and infrastructure operators early, structured facts, including uncertainty
First day
Restricted operations
- •Move to restricted operations where safe and publish time-stamped customer guidance
- •Protect cash and essential payment access
- •Coordinate liquidity actions and settlement priorities with relevant authorities
- •Preserve original logs, API calls, configuration changes and approval records; do not treat model summaries as source evidence
Recovery period
Trusted restoration
- •Rebuild from known-good assets and revoke and reissue affected credentials
- •Validate model and tool versions, test data lineage and reconcile all transactions
- •Obtain independent approval before restoring normal agent authority
- •Quantify customer harm, loss allocation, backlogs and control failures; retest remediations
LUMINAIRE and CABIER Consulting are affiliated platforms. The simulator is an educational adaptation, not an independent audit.