Financial Services

Posture

Financial services is the vertical with the deepest pre-existing model-risk discipline. Banks, insurers, and asset managers have run model-validation, model-inventory, and model-risk-management programmes for decades: under SR 11-7 in the United States, PRA SS1/23 in the United Kingdom, and EU model-risk frameworks across capital, liquidity, and conduct dimensions. The arrival of frontier-model AI has not displaced this discipline; it has stress-tested it.

BtSavvy's posture is to extend the existing model-risk discipline rather than bypass it. Frontier-model and traditional-ML systems alike sit inside the bank's model inventory; both inherit model-risk lifecycle controls; both produce conformance evidence the model-risk function consumes directly. FORGE is structured to feed existing model-risk operating models rather than require a parallel one.

Regulatory posture

EU AI Act: many financial-services AI systems fall under Annex III high-risk classifications, particularly creditworthiness scoring (Annex III, point 5(b)), risk assessment in life and health insurance pricing (point 5(c)), and AI used in employment and worker management.

Model risk management — SR 11-7 (US), PRA SS1/23 (UK), ECB guide to internal models (EU, for regulatory-capital models): model-validation, model-inventory, and model-lifecycle discipline applied to AI/ML alongside traditional statistical models.

EU DORA: operational-resilience obligations including ICT third-party risk management, applicable to AI consumed from third-party providers.

MAS, FCA, and other conduct-regulator AI guidance: conduct-of-business AI expectations including consumer-fairness, explainability, and accountability (MAS FEAT principles, FCA AI discussion papers, others).

UK GDPR / EU GDPR plus consumer-credit provisions: Article 22 automated-decision-making rights apply to creditworthiness and similar consequential AI decisions. Basel III/IV (where applicable): capital implications for model-driven decisions. ISO/IEC 42001: AI Management System standard.

Which FORGE services carry the most weight

Assurance. The load-bearing service: composing continuous conformance evidence across the applicable regimes: EU AI Act, model-risk (SR 11-7 / SS1/23), and conduct-regulator expectations.

Data Foundation. Financial-services data is special-category-rich, regulator-traceable, and consent-bound: transactional, market, customer, and regulatory data brought into agent-ready form under lineage discipline the model-risk function can consume directly.

Strategy & Architecture. Typically establishing the model-risk integration architecture, regulator-engagement posture, and deployment-and-monitoring framework.

Build. Application engineering: risk-scoring models, fraud-detection systems, AML and customer-due-diligence systems, customer-facing assistants, market-making and execution systems.

How FORGE works in financial services

Factory produces accelerators calibrated against financial-services failure modes: credit-model performance regression by demographic subgroup, fraud-model false-positive cost monitoring, AML-model alert-quality monitoring.

Observability emits signals consumable directly by the model-risk function (calibration-skew, performance-by-subgroup, conceptual-drift indicators) at the cadence and resolution model-risk validation requires.

Review produces signed gate logs against the gates the model-risk function expects: model registration, validation, promotion, periodic re-validation, retirement. The model-risk function consumes the gate log as evidence directly rather than reconstructing it.

Governance and Trust encodes model-risk-tier classification and model-validation cadence as policy-as-code gates; model retirement is a signed promotion/retirement gate rather than procedural narrative.

Expert Intelligence combines frontier-model competence with financial-services domain expertise (risk-modelling, market-data reasoning, regulatory-text retrieval, fraud-pattern recognition) with domain fine-tuning or retrieval against financial data products and regulatory corpora.

Use-case sketches

A creditworthiness AI deployment with EU AI Act high-risk discipline. A frontier-model-augmented or traditional-ML credit-decision system with full EU AI Act Article 14 human-oversight controls enforced as a promotion gate, a GDPR Article 22 challenge mechanism, and model-risk validation evidence emitted continuously by the delivery system.

An AML / customer-due-diligence system under DORA discipline. An AML system with full ICT third-party risk management for any externally hosted model components, recorded at every alert disposition, and full traceability from transaction through model through investigator decision.

A frontier-model customer assistant with FCA conduct-of-business conformance evidence. A customer-facing AI assistant with FCA-aligned conduct-of-business controls (consumer-fairness monitoring, explainability surfaces, complaint-handling integration) running as live SRE signals rather than periodic compliance assessment.

Insights cross-links