RESEARCH

Peer-reviewed methodology. Published findings.

Research published on SSRN. Peer review: Journal of Financial Stability. WBSO deep-tech R&D (2026). All findings are reproducible from the published methodology.

1

Topological Governance of Autonomous AI Agents in Regulated Financial Services: An Empirical Analysis with Formal Hypothesis Testing

Collins (2026a) · Published, SSRN

95.5% halt-absent across 49,231 agents. SPOF paradox: institutions that invested most built the most fragile architecture. Coase Inversion: governance concentration exceeds vendor concentration.

N=263 institutions, 49,231 agents, 162,399 edges, 45 countries. Weight perturbation rank correlation 0.988. Split-half reliability 0.988 (Spearman-Brown 0.994).

AI agent governance, halt mechanism, DORA compliance, network topology, governance concentration, SPOF, financial regulation, autonomous AI oversight

ssrn.com/abstract=6470098
2

Cascade Without Containment

Collins (2026b) · Under review, Journal of Financial Stability

100% of institutions return UNCONTAINED cascade verdict. Mean blast radius 29.8% of agent population (range 8.1% to 65.4%). Governance failure propagates through 1 to 3 hops.

10,000 Monte Carlo simulations. Mean applicable regulators per institution: 7.2. 87% score above 3.0/4.0 on compound severity.

cascade simulation, contagion, blast radius, Monte Carlo, governance failure propagation, systemic risk, financial stability

Under peer review
3

Credit Ratings Cannot Measure What They Cannot See: The Case for an Independent AI Governance Rating in Financial Services

Collins (2026c) · Published, SSRN

S&P correlation with governance: r = 0.042, R² = 0.002. Four South African banks with identical BB rating show 8× governance gap. Within the AA band: 41× governance gap.

S&P N=179, Moody's N=159, Fitch N=143. Bootstrap 95% CI: [-0.160, 0.219]. LOO max influence: 0.034.

credit ratings, AI governance rating, MAR, S&P, Moody's, Fitch, governance benchmark, financial services, rating blind spot

ssrn.com/abstract=6524438
4

Systemic Governance Risk in AI Agent Networks: Cross-Institutional Contagion and Topological Vulnerability

Collins (2026d) · Published, SSRN

GCI 3.7×: governance concentration exceeds vendor concentration in 82% of institutions. 52,652 pairwise cross-institutional links. 35.8% connect two institutions both scoring below 5.0.

Top 5 governance hubs are all committees, not technology platforms. System-level HHI: 360 (unconcentrated). Per-institution HHI: 2,689 (concentrated).

systemic risk, cross-institutional contagion, vendor concentration, governance concentration index, topological vulnerability, AI agent networks

ssrn.com/abstract=6535599
ROBUSTNESS
0.988
Weight perturbation rank correlation
0.988
Split-half reliability (Spearman-Brown 0.994)
0.929
Convergent validity (max r)
0.034
LOO max influence