Life sciences · Preprint

Cyber-Financial Contagion: Modeling the Propagation of an AI Vendor Compromise Through the Banking System

arXiv · September 9, 2026

Hypothesis-Generating Why this matters to you Expert discussion

Raises a question worth testing. It does not answer one.

AI Summary

This is an unpublished computational study that models how a compromise of a shared AI vendor could propagate financial losses through the banking system via a four-layer heterogeneous network. The authors propose two tools (CFC-Prop, a stochastic epidemic model, and CFC-GNN, an early-warning classifier) and report model performance on synthetic data, but provide no empirical validation, real incident data, or peer review.

What was studied

Computational modeling and machine learning study using synthetic network data. Intervention: Simulated compromise of AI vendor within synthetic banking network; CFC-Prop contagion propagation model. Compared with: Four machine learning baselines for early-warning classification.

Key findings

Synthetic network comprises 60 AI vendors, 220 banks, approximately 2,500 vendor-bank service edges, and 1,400 interbank exposures CFC-Prop model reproduces heavy-tailed loss distributions and sharp dependence on patch latency consistent with prior cyber-financial evidence CFC-GNN early-warning model reaches AUROC 0.82 and AUPRC 0.60 on vendor-side incident telemetry and graph structure

Safety

Safety was not reported in the material analysed. Check the source before drawing any conclusion about harm.

Who this may help

The source did not state who this applies to in practice.

Plain-language summary

A modeling and simulation study that raises important questions about AI vendor concentration risk in banking, but does not test real-world outcomes or provide empirical validation beyond synthetic data.

Study design

As stated by the source record.

MethodsAuthors constructed a four-layer heterogeneous network coupling AI vendors, financial institutions, interbank exposures, and customer accounts. They developed CFC-Prop, a stochastic epidemic-and-clearing model, to simulate contagion on this network. They trained CFC-GNN, a graph neural network classifier, on vendor-side incident telemetry and network structure to predict high-cascade-risk vendors. Performance metrics (AUROC, AUPRC, calibration error) were computed on the synthetic dataset. Code, synthetic data, and reproducible scripts were released.

DesignComputational modeling and machine learning study using synthetic network data

InterventionSimulated compromise of AI vendor within synthetic banking network; CFC-Prop contagion propagation model

ComparatorFour machine learning baselines for early-warning classification

Reported figures

Quoted from the source exactly as published.

Vendors in synthetic network60

Banks in synthetic network220

Vendor-bank service edges2500

Interbank exposures1400

CFC-GNN AUROC0.82

CFC-GNN AUPRC0.60

Caveats

This is a preprint and has not undergone peer review
All data are synthetic and no validation against real banking crises, actual vendor incidents, or observed financial losses is reported
Model validation and calibration approach not described; external test set performance not reported
No comparison of CFC-GNN against actual vendor incident prediction in operational settings

Evidence assessment

Graded across the dimensions that decide whether you should act, each from what the source actually supports. There is no single score, and where a dimension was not assessed it says so.

Study design

limited

Synthetic network model without real-world data validation or empirical trial.

Directness

limited

Addresses a plausible mechanism but uses hypothetical network and synthetic scenarios, not observed events.

Consistency

moderate

Model outputs reported as consistent with prior cyber-financial evidence, but no direct comparison studies cited.

Risk of bias

insufficient

Synthetic data design and validation approach not described; external validation not reported.

Sample adequacy

moderate

Synthetic network includes 60 vendors, 220 banks, 2500 vendor-bank edges, and 1400 interbank exposures.

Source abstract

The banking system now depends on a small set of shared artificial intelligence vendors for fraud screening, credit decisioning, anti-money-laundering triage, customer analytics, and internal decision support. This paper studies how a compromise inside one of those vendors can propagate along a chain of operational, informational, and financial linkages until it triggers losses that look, from the outside, like a classical banking crisis. We build a four-layer heterogeneous network that couples AI vendors, financial institutions, interbank exposures, and customer accounts, and we propose CFC-Prop, a stochastic epidemic-and-clearing model that runs on that network. On a synthetic dataset with 60 vendors, 220 banks, roughly 2,500 vendor-bank service edges, and 1,400 interbank exposures, CFC-Prop reproduces the heavy-tailed loss distributions and the sharp dependence on patch latency that are consistent with prior cyber-financial evidence. We also train an early-warning model, CFC-GNN, that uses vendor-side incident telemetry and graph structure to flag high-cascade-risk vendors before impact. Across four baselines the proposed model reaches AUROC 0.82 and AUPRC 0.60 while keeping calibration errors bounded. We release the full code, synthetic data, and reproducible scripts. The results argue that cyber concentration among AI vendors is a first-order financial-stability problem and give supervisors a concrete quantitative tool for reasoning about it.

Follow from this

Taken from the source record, never inferred. Follow any of these and new work involving them reaches your briefing.

Alex Leytes