HomeAI OfferingsAuditCase studiesAboutContact
COMPLIANCE

RBI FREE-AI explained: what India's AI banking regulation actually requires

Plain-English breakdown of RBI FREE-AI — the 7 Sutras, the 6 pillars, Annex V and VI templates, and the 4-phase rollout. What NBFCs and banks need in place before Phase 3 mandatory audit.

The RBI's Framework for Responsible and Ethical Enablement of AI (FREE-AI) is the first detailed, AI-specific regulation in Indian financial services. It does not replace existing risk management circulars — it sits on top of them. If you are an NBFC, bank, or payment system operator deploying any AI feature in 2025, FREE-AI is the document you need to read.

The 7 Sutras

FREE-AI organises its principles into seven Sutras: Trust, People-First, Innovation Under Risk-Based Regulation, Fairness and Equity, Accountability, Understandable by Design, and Safety, Resilience and Sustainability. These are not slogans — each maps to specific obligations later in the document.

"Understandable by Design" in particular sets a high bar. It is not enough to say a model is explainable; the explanation must be intelligible to the affected customer in the language they were served in.

The 6 pillars

Governance, Risk Management, Model Lifecycle, Data Management, Third-Party Management, and Audit and Assurance. Each pillar is laid out as a set of operational requirements, not principles. Governance requires a named officer; Model Lifecycle requires documented model cards; Audit and Assurance requires evidence that an independent function can review the system.

Annex V and Annex VI

Annex V is a model card template — the artefact you produce per model in production. Annex VI is the AI use-case register — the inventory of every place AI is used in your regulated entity, classified by risk tier.

Both are auditable artefacts. Both need to be in place before Phase 3 begins. "We use OpenAI" is not a model card.

The 4-phase rollout

Phase 1 — self-assessment against FREE-AI and gap remediation planning. Phase 2 — implementation of governance, model registry, and DPIA-equivalent process for AI. Phase 3 — mandatory audit by qualified independent assessors. Phase 4 — supervisory review by the RBI and ongoing reporting.

Most regulated entities are currently in Phase 1 or early Phase 2. The earlier you stand up the governance and registry, the cheaper Phase 3 becomes — because the auditor is looking for evidence, not intent.

What "AI use" actually triggers FREE-AI

Anything that materially influences a customer-facing decision: credit underwriting, fraud detection, suitability scoring for investment, KYC risk grading, complaint routing, chatbot responses that quote rates or terms. "It is a third-party API" is not a defence — third-party AI is explicitly in scope under the Third-Party Management pillar.

How to get from zero to Phase 3 ready

Inventory every AI use case (Annex VI). Build a model card for each (Annex V). Stand up a governance forum that meets and minutes. Establish a DPIA-equivalent for new AI projects. Build a third-party AI register and renegotiate where needed. Engage an independent reviewer to dry-run the Phase 3 audit before the deadline.

Our compliance programme runs this end-to-end for NBFCs and banks, mapped against DPDP Act in parallel so the same control evidence serves both regulators.

MORE FROM THE BLOG

Keep reading.

AI SECURITY10 min

The five security holes every vibe-coded app ships with

Lovable, Bolt, Cursor, and v0 ship the same five vulnerabilities every single time. Specific, reproducible, and exactly what an attacker looks for first.

Read the post →
COMPLIANCE8 min

DPDP Act 2023 and AI: what "Data Fiduciary" actually means for your startup

Plain-English walkthrough of when an AI system makes you a Data Fiduciary under DPDP Act 2023, what the obligations are, and what the ₹250 crore penalty actually attaches to.

Read the post →
AI SECURITY12 min

Why RAG pipelines leak — and how to test yours

Cross-tenant retrieval leakage is the most under-diagnosed vulnerability in production RAG systems. Where namespace isolation breaks, what to test, and how to fix it at the architecture level.

Read the post →
BOOK A DISCOVERY CALL

30 minutes. No pitch.