Seven Sutras, One Platform:
How Indian Banks Can Operationalise RBI's AI Principles

The Seven Sutras at a Glance

The FREE-AI committee structured seven interconnected values: Trust is the Foundation, People First, Innovation over Restraint, Fairness and Equity, Accountability, Understandable by Design, and Safety, Resilience & Sustainability. Together they articulate a vision of AI that is trustworthy, human-centred, fair, accountable, explainable, and resilient.

The Operationalisation Gap

PwC's 2025 Responsible AI survey found that while 60% of executives acknowledged responsible AI's contribution to ROI, nearly half reported difficulty translating principles into operational processes. Deloitte's 2026 Banking Outlook found AI initiatives constrained by fragmented data foundations, legacy systems, and mounting compliance demands — with many confined to isolated proofs of concept.

Sutra One — Trust is the Foundation

End-to-end data lineage and provenance tracking across federated data sources. When an AI agent produces a credit recommendation, the institution must trace every data input — from which source system, when it was accessed, how it was transformed — back to origin. A federated architecture that queries data in place preserves the provenance chain intact.

Sutra Two — People First

Human-in-the-loop governance with configurable escalation thresholds. AI agents must operate within defined guardrails that automatically escalate decisions to human reviewers when confidence scores fall below thresholds, when the decision impacts a vulnerable population, or when regulatory sensitivity is high.

Sutra Four — Fairness and Equity

Federated access to diverse data sources with embedded bias detection. Traditional credit models reliant on bureau scores systematically exclude thin-file borrowers. A federated architecture enriches credit assessments by accessing alternative signals — UPI transaction patterns, GST filings, Account Aggregator data — without centralising sensitive information.

Sutra Five — Accountability

Immutable decision audit trails that log every agent action, every data input consumed, every rule applied, and every output generated. When a regulator asks "why was this borrower declined?", the institution must reconstruct the complete decision chain within minutes, not days.

Sutra Six — Understandable by Design

Decision-level explainability, not just model-level interpretability. A SHAP value chart means nothing to a regulator asking why a particular MSME was declined. The platform must translate model outputs into natural language explanations — identifying key factors, their relative weights, and the policy rules applied. This is the distinction between interpretability and explainability.

Why Point Solutions Will Not Work

The temptation is to address each sutra with a dedicated tool. This approach creates precisely the kind of fragmented governance landscape the FREE-AI framework was designed to prevent. Gartner's Q2 2025 survey found organisations deploying dedicated AI governance platforms were 3.4× more likely to achieve high governance effectiveness than those relying on fragmented approaches.

Responsible AI demands a phased approach that balances innovation, inclusion, and stability.

— T. Rabi Sankar, RBI Deputy Governor, Global Fintech Festival 2025