← Back to tools

Tool 06 · TOOL 06

AI Model Accumulation Assessor

Quantify the logical concentration of your AI insurance portfolio by foundation model. Identify silent exposures from cross-model dependencies.

Parameters
Portfolio profile
Exposed lines of business
AI-exposed premium volume (M€)100 M€
0250500 M€
Number of active AI policies
Foundation model inventorymax. 6 models
Foundation model 1
Foundation model
Cloud provider
Primary use
% book exposed30 %
Dependency depth
% of models at the same cloud provider40 %
Technical integration depth2/5 · Light integration
API purPropriétaire
AI liability clauses in policies
EU AI Act compliance verified by insured
AI governance maturity

Accumulation Index

62/100
050100
High

Significant accumulation, action recommended

Accumulation formula: +45
Provider conc.0
RI clauses+5
EU AI Act+8
Integration depth+4
Governance0

Model × Lines of Business Heatmap

P&CCyber
GPT-4o
Low
Medium
High

Model PML by foundation model (M€)

GPT-4o24.0 M€
24% of the AI book · Underwriting · OpenAI / Microsoft Azure. This model carries 100% of aggregate PML.

EU AI Act Flags

High risk
GPT-4o
High-capability model in critical use. EU AI Act Article 6 potentially applicable.

Top 3 insights

Critical concentration: 100% of the AI book sits with OpenAI / Microsoft Azure. An unplanned outage triggers a correlated loss across the full perimeter.

HHI of 10000: high concentration by OECD thresholds. Systemic accumulation risk is non-diversifiable under the current configuration.

1 model(s) classified EU AI Act "high risk" (GPT-4o) with no compliance check. Exposure to Article 6 and DORA sanctions before August 2026.

HHI Concentration

10000High concentration

OECD thresholds: < 1,500 diversified, 1,500-2,500 moderate, > 2,500 high.

OpenAI / Microsoft Azure100%

Correlated outage scenario

If OpenAI / Microsoft Azure is unavailable 24h

Dominant providerOpenAI / Microsoft Azure
Direct exposure30 M€ (100%)
Correlation rate50%
Estimated correlated loss15 M€

Estimate based on declared integration depth (24h outage hypothesis).

Methodology & disclaimer

Accumulation formula: score = min(100, sum(% book x usage coeff) + concentration and compliance penalties). Coefficients: Underwriting 1.5 / Claims 1.4 / Compliance 1.3 / Risk 1.2 / Service 1.0.

Model PML: PML = (% book / 100) x premium volume x usage severity factor. Factors: Underwriting 0.8 / Claims 0.7 / Compliance 0.5 / Risk 0.6 / Service 0.3.

EU AI Act:Indicative risk levels based on model capabilities and declared use. Not a substitute for legal analysis.

HHI : Herfindahl-Hirschman Index computed on provider exposure shares. OECD thresholds: < 1,500 diversified, 1,500-2,500 moderate, > 2,500 concentrated.

Integration depth: Scale 1-5 representing model substitutability: 1 = stateless API (replacement in hours), 5 = internalized proprietary model (migration > 30 days). Score impact: +4 pts per level.

Results depend on entered parameters and do not constitute a certified accumulation risk assessment.

Illustrative research tool, not certified. AlgoPolis, 2026.