Modular AI Governance for Finance – Access Partnership
Building systems that are inclusive, resilient, and interoperable
Artificial intelligence is now embedded in nearly every layer of finance, from credit underwriting and trading to fraud detection and supervision. But as adoption accelerates, governance frameworks have not kept pace.
To avoid fragmentation and risk, AI governance in finance must be modular, additive, and mission-driven – not monolithic or reactive. Its objectives should be clear: to make finance inclusive, resilient, and interoperable.
This week, alongside the IMF–World Bank Fall Meetings in Washington, DC, Access Partnership joined the Atlantic Council’s AI x Finance panel with industry and policy experts to examine how AI can advance these goals. The panel covered the role of the private sector, the importance of international standards, and how best to close governance gaps. The question now is how to translate these insights into a workable governance framework.
Global momentum: FSB and BIS take the lead
The Financial Stability Board (FSB) recently released its report, Monitoring Adoption of Artificial Intelligence and Related Vulnerabilities in the Financial Sector (October 2025).
It highlights three urgent needs: closing data gaps, addressing third-party concentration, and harmonising reporting on AI use in financial systems.
Meanwhile, the Bank for International Settlements (BIS) advanced several complementary initiatives:
- Project AISE (AI Supervisory Enhancer) – a prototype virtual assistant for supervisors to inspect models, detect anomalies, and benchmark algorithmic behaviour across institutions.
- Project Noor – a joint effort with the UK Financial Conduct Authority (FCA) and Hong Kong Monetary Authority (MA) to build explainability and model-audit tooling for AI in financial services.
- Project Meridian FX – a cross-border settlement experiment linking real-time gross settlement (RTGS) and distributed ledger technology (DLT) systems for atomic foreign-exchange settlement, proving that interoperability can be achieved without forcing everyone onto the same ledger.
Together, these projects outline a new architecture for AI governance – distributed, data-rich, and tool-assisted.
Regional proof points: modular governance in action
India’s Digital Public Infrastructure
India’s Digital Public Infrastructure (DPI), built around Aadhaar, UPI, and the Account Aggregator framework, shows how modular building blocks can unlock inclusion.
Each layer (identity, payments, consented data-sharing) works independently yet interoperates through open APIs. It enables AI-driven credit scoring and “credit at payout” for SMEs and gig workers, proving that digital trust layers can scale safely when governance is modular.
MAS FEAT and Veritas Toolkits
The Monetary Authority of Singapore’s FEAT principles and Veritas Toolkit demonstrate additive governance. Banks start with fairness and explainability checks, then integrate bias testing and accountability dashboards as maturity grows. It’s modular by design, with governance as an evolving stack, rather than a one-off compliance exercise.
BIS Project Nexus: interoperability at scale
Project Nexus connects national instant-payment systems (PayNow, UPI, DuitNow, PromptPay) through a shared messaging gateway. It’s a technical and governance blueprint for AI-ready corridors, where transactions, identity, and compliance proofs can move seamlessly across borders.
A proposed modular governance stack
Access Partnership is proposing a governance stack to complement the financial and technical stack. Each module can be adopted incrementally, focusing on additive governance that grows with institutional capacity.
| Module | Purpose | Interfaces |
| AI Inventory & Metadata Catalogue | Central register of models, lineage, and purpose | Audit & supervisory access |
| Explainability & Model Cards | Standardised documentation of logic and limits | Automated reporting |
| Bias & Fairness Testing | Segment-level error rates and drift alerts | Dashboards, retraining triggers |
| Resilience & Stress Testing | Shock simulation, fail-over models | Scenario libraries |
| Third-Party Risk | Transparency for vendor and model providers | Contractual disclosures |
| Interoperable Compliance Layer | Schema for identity, attestations, and receipts (vLEI + ISO 20022) | API-based proofs |
| Governance Dashboard | Quarterly indicators (adoption, incidents, concentration) | Public or regulator view |
From principles to practice: how it can happen
Pilot cross-border corridors
Test interoperable compliance, model audit exchange, and synchronised receipts using BIS Nexus or Meridian FX frameworks.
Adopt minimal AI indicators
Align with FSB recommendations, publish metrics on AI model use, AI impact (such as approval rate parity), third-party reliance, and failure rates.
Deploy supervisory tooling
Use AISE or Noor prototypes to augment supervisors with explainability and anomaly detection.
Expand modular adoption
Gradually introduce fairness testing, resilience simulation, and vendor-exit planning.
Cooperate on standards globally
Through BIS, FSB, and ISO forums, standardise attestation schemas and mutual recognition mechanisms so AI models can be audited once and trusted globally.
Why it matters
This modular approach transforms governance from a compliance checklist into an enabler of digital trust, by focusing on making financial systems:
- Inclusive – Bias and explainability modules make AI lending and onboarding fairer.
- Resilient – Stress-testing modules ensure systems can absorb correlated model failures.
- Interoperable – Standardised receipts and identity proofs let trusted data flow securely across borders.
Looking ahead
AI governance for finance will define the next decade of financial stability and inclusion.
If we design it to be modular, additive, and interoperable, it will scale like the systems it oversees. As BIS, FSB, and regional innovators from Singapore to India have shown, this future is already taking shape, one building block at a time.
Watch Abhineet’s full discussion at the Atlantic Council’s AI x Finance panel.
