India AI & sector rulesUpdated 22 November 202610 min read

RBI FREE-AI Framework: What Banks & Fintechs Must Do

What the RBI FREE-AI committee report asks of banks, NBFCs and fintechs, what is and is not binding yet, and how it maps onto DPDP obligations from May 2027.

By Suhas Bhairav · Reviewed against the DPDP Rules, 2025 (G.S.R. 846(E), 13 Nov 2025)

Full name
Framework for Responsible and Ethical Enablement of AI
Issued by
Committee constituted by RBI; report released August 2025
Committee chair
Prof. Pushpak Bhattacharyya
Structure
7 sutras, 26 recommendations, 6 pillars
Legal status
Recommendations; binding only via RBI directions or circulars
Related DPDP date
Core fiduciary obligations from 13 May 2027

Quick answer

RBI's FREE-AI (Framework for Responsible and Ethical Enablement of AI) is a committee report released in August 2025 with 7 sutras and 26 recommendations across six pillars. It asks regulated entities for a board-approved AI policy, an AI inventory, risk-based classification, customer disclosure, grievance escalation, AI incident reporting and model audits. It becomes binding only when RBI issues directions or circulars implementing it.

  • FREE-AI is a committee report, not a direction; its recommendations bind regulated entities only once RBI issues implementing directions or circulars.
  • The framework has 7 sutras and 26 recommendations grouped under six pillars: infrastructure, policy, capacity, governance, protection and assurance.
  • The core operational asks are a board-approved AI policy, an AI inventory, risk classification, customer disclosure, human escalation and AI incident reporting.
  • FREE-AI proposes a tolerant supervisory approach to first-time AI errors where adequate safeguards were in place.
  • DPDP security, breach and retention duties apply to the same AI systems from 13 May 2027 regardless of RBI's timeline.

Who this is for: Banks and small finance banks, NBFCs and housing finance companies, Payment system operators and fintechs, Chief Risk Officers and CISOs, Model risk and data science teams

What is the RBI FREE-AI framework?

FREE-AI stands for the Framework for Responsible and Ethical Enablement of AI. It is the report of a committee set up by the Reserve Bank of India and chaired by Prof. Pushpak Bhattacharyya, released in August 2025.

The report does two things at once. It encourages AI adoption in Indian finance, and it sets out what responsible adoption should look like for regulated entities (REs) such as banks, NBFCs and payment system operators.

Structure at a glance

  • 7 sutras — guiding principles that frame the rest of the report.
  • 26 recommendations — specific actions for RBI, the industry and individual REs.
  • 6 pillars — infrastructure, policy, capacity, governance, protection and assurance.

The first three pillars (infrastructure, policy, capacity) are about enabling AI: shared infrastructure, an AI innovation sandbox, and skills. The last three (governance, protection, assurance) are about managing risk: board oversight, consumer protection, cybersecurity, audits and validation.

Is it binding?

Not on its own. A committee report states recommendations. They become obligations for REs only when RBI issues directions, master directions or circulars that implement them. Before relying on any specific requirement or deadline, check the latest circulars on rbi.org.in.

That said, supervisors tend to ask about published expectations early. Treating FREE-AI as the expected direction of travel is the practical position.

What FREE-AI expects from banks, NBFCs and fintechs

Stripped to operational terms, the report's asks for regulated entities fall into ten items.

ExpectationWhat it means in practice
Board-approved AI policyA written policy, owned by the board, covering AI use, risk appetite and accountability
AI inventoryA repository of every model and AI use case, internal and vendor-supplied
Risk-based classificationEach use case tiered by impact on customers and the institution
Governance oversightNamed owners, committee reporting and approval gates
Customer disclosureTelling customers when they are interacting with AI
Grievance and human escalationA route from an AI interaction to a human who can resolve it
AI incident reportingCapturing and reporting AI-specific failures, not just cyber incidents
Audits and model validationIndependent testing before and after deployment
Cybersecurity for AIControls for prompt injection, model theft, data poisoning and similar threats
Tolerant supervisionFirst-time errors treated leniently where safeguards existed

The last item matters. The report proposes a supervisory stance that does not punish a first-time AI error harshly if the RE can show it had safeguards in place. That makes documentation a direct risk reducer: your inventory, validation reports and incident logs are your evidence.

For a copy-ready policy structure built on these items, see the board AI policy template.

Infrastructure, sandbox and capacity: the enabling side

FREE-AI is not only a risk document. Half of its pillars aim to make AI easier for the financial sector to adopt.

Shared infrastructure

The report proposes shared infrastructure for the sector, so that smaller REs (cooperative banks, smaller NBFCs) are not locked out by the cost of compute, data and models. How that infrastructure would be built, funded and governed is a matter for RBI and the ecosystem to settle; watch for announcements rather than assuming a design.

AI innovation sandbox

The report also proposes an AI innovation sandbox, letting REs and fintechs test AI products under supervisory oversight before full rollout. If you are building credit, fraud or customer-service models, a sandbox route can shorten the path to supervisory comfort. Check RBI's site for whether a dedicated AI sandbox cohort has been opened and on what terms.

Capacity building

Boards, senior management and supervisors are expected to understand AI well enough to oversee it. In practice this means board-level training, a named executive accountable for AI, and technical staff who can explain model behaviour in plain language.

Why this matters to compliance teams

Enablement and risk are linked. An RE that uses the sandbox, keeps a clean inventory and trains its board is building the same evidence base that the governance and assurance pillars ask for.

How to implement FREE-AI controls in real systems

Most REs already run model risk management for credit scorecards. FREE-AI extends that discipline to generative AI, vendor APIs and agents.

Build the AI inventory first

Create one register with a row per use case. Minimum fields:

  • use case name, business owner and technical owner;
  • model type (in-house ML, foundation model API, vendor product);
  • personal data categories processed and the DPDP ground (consent or section 7 legitimate use);
  • risk tier and the reason for it;
  • customer-facing or internal;
  • validation date, next review date and incident history.

Discover shadow AI by scanning outbound traffic for model API domains and reviewing procurement records.

Wire disclosure and escalation into the product

For customer-facing chatbots and voice bots, show an AI disclosure at the start of the interaction and offer a visible "talk to a person" path. Log every escalation with a ticket ID so the grievance trail is auditable.

Treat AI incidents as a category

Define an AI incident (harmful output, biased decision pattern, data leakage through a model, prompt injection that triggered an action). Route them through the same incident desk as cyber events, with an AI tag. Some will also be personal data breaches under DPDP and cyber incidents under the CERT-In directions.

Validate and monitor

Run pre-deployment validation (accuracy, fairness across customer segments, robustness, explainability) and post-deployment drift monitoring. Store reports against the inventory row.

How FREE-AI interacts with the DPDP Act and Rules

FREE-AI governs how REs use AI. The DPDP Act governs personal data inside those systems. Both apply to the same model pipeline, and the DPDP obligations have fixed dates.

FREE-AI itemDPDP obligation on the same systemApplies from
Cybersecurity for AIRule 6 reasonable security safeguards: encryption, access control, logging, backups13 May 2027
AI incident reportingRule 7 breach intimation to each Data Principal and to the Board within 72 hours13 May 2027
AI inventoryData mapping needed for notices (Rule 3) and rights requests (Rule 14)13 May 2027
Customer disclosureRule 3 notice: itemised data and specified purpose13 May 2027
Audits and validationRule 13 DPIA, audit and algorithmic due diligence, if notified as an SDFOnce notified

Three points to note:

  1. Logs. Rule 6(1)(e) and Rule 8(3) require at least one year of logs and associated data. Your AI inference logs are in scope.
  2. No materiality threshold. Every personal data breach is reportable under Rule 7, including leakage through a model output.
  3. Sectoral law prevails. Section 16(2) of the DPDP Act keeps stricter rules such as RBI's payment data localisation in force.

For the wider sector view, see DPDP for banks, NBFCs and fintechs.

Common mistakes when responding to FREE-AI

  • Waiting for a circular before doing anything. Inventory and classification take months in a large bank. Starting now costs little and pays off under any final direction.
  • Inventorying only in-house models. Vendor chatbots, CRM copilots and employee use of public LLMs are AI use cases too. FREE-AI's inventory idea covers them.
  • Treating disclosure as a footer. Customers should know they are dealing with AI at the point of interaction, not in terms and conditions.
  • No human route. A chatbot that cannot hand off to a person fails both the FREE-AI escalation ask and DPDP grievance expectations under section 8(10) and Rule 14(3).
  • Ignoring vendors. Outsourcing an AI function does not outsource accountability. Contracts need validation access, incident notification and Rule 6(1)(f) security clauses.
  • Quoting unverified numbers. Do not cite deadlines or thresholds attributed to FREE-AI unless you have read them in an RBI direction. The report is a recommendation set.

What to do before 13 May 2027

Finish the inventory, approve the board policy, implement disclosure and escalation for customer-facing AI, and align AI incident handling with the DPDP breach workflow. That sequence satisfies the likely shape of RBI directions and the fixed DPDP dates together.

Further reading on suhasbhairav.com

Before 13 May 2027

Compliance checklist

  1. Read the FREE-AI report and track RBI circulars

    Assign a compliance owner to monitor rbi.org.in for directions implementing FREE-AI recommendations.

    RBI FREE-AI report

  2. Build an AI inventory

    Register every model and AI use case, including vendor and employee tools, with owners and risk tiers.

    FREE-AI governance pillar

  3. Approve a board AI policy

    Table a policy covering scope, risk appetite, roles, approvals and reporting to the board.

    FREE-AI policy pillar

  4. Classify use cases by risk

    Tier each use case by customer impact, autonomy and data sensitivity; apply controls by tier.

    FREE-AI governance pillar

  5. Disclose AI to customers

    Show AI disclosure at the start of chatbot, voice and email interactions handled by AI.

    FREE-AI protection pillar

  6. Provide human escalation

    Give customers a visible path to a human and log escalations for grievance audit.

    DPDP s.8(10), Rule 14(3)

  7. Define AI incidents

    Add an AI incident category to the incident desk and map it to DPDP breach and CERT-In reporting.

    DPDP Rule 7

  8. Validate and monitor models

    Run pre-deployment validation and ongoing drift and fairness monitoring; file reports in the inventory.

    FREE-AI assurance pillar

  9. Retain AI logs for one year

    Keep inference and access logs for at least one year to meet DPDP Rule 6(1)(e) and Rule 8(3).

    DPDP Rule 6(1)(e), Rule 8(3)

Key dates

  1. DPDP Rules notified; Board-related rules in force

  2. DPDP core obligations (security, breach, retention, rights) apply

Answers

Frequently asked questions

Is RBI FREE-AI mandatory for banks?

Not by itself. FREE-AI is a committee report released in August 2025. Its 26 recommendations become mandatory for banks, NBFCs and other regulated entities only when RBI issues directions or circulars that implement them. Check rbi.org.in for current status. In practice, supervisors use published expectations as a reference, so preparing early is sensible.

What are the six pillars of FREE-AI?

The six pillars are infrastructure, policy, capacity, governance, protection and assurance. The first three focus on enabling AI adoption through shared infrastructure, an innovation sandbox and skills. The last three focus on managing risk through board oversight, consumer protection, cybersecurity, audits and model validation.

Does FREE-AI apply to NBFCs and fintechs?

FREE-AI is addressed to RBI-regulated entities, which include banks, NBFCs and regulated payment system operators. Fintechs that are not themselves regulated are usually affected indirectly, because their bank or NBFC partners will push FREE-AI style controls into contracts, due diligence and audits.

Do banks have to tell customers they are talking to AI?

FREE-AI recommends that regulated entities disclose to customers when they interact with AI, and provide grievance redress with escalation to a human. Whether this is a binding rule depends on RBI issuing a direction. Separately, DPDP Rule 3 already requires a clear notice of the personal data and purposes involved from 13 May 2027.

How does FREE-AI treat AI mistakes?

The report proposes a tolerant supervisory approach to first-time errors where the regulated entity had adequate safeguards in place. That makes documentation valuable: an inventory, validation reports, incident logs and board approvals are the evidence that safeguards existed when something went wrong.

Is an AI incident the same as a DPDP data breach?

Not always. An AI incident might be a biased decision pattern with no data exposure. But if a model leaks personal data or an attacker uses an AI system to access it, that is a personal data breach under DPDP, reportable to each affected person and to the Board, with a detailed report within 72 hours.

When should a bank involve legal counsel on FREE-AI?

Involve counsel when RBI issues a direction implementing FREE-AI, when drafting vendor contracts for AI services, and when an AI incident may also be a personal data breach or a reportable cyber incident. Those are the points where obligations become legally specific and time-bound.

Legal references and sources

  • RBI FREE-AI Committee Report — Recommendations across six pillars (official text)
  • Digital Personal Data Protection Rules, 2025 — Rule 6 (security safeguards) (official text)
  • Digital Personal Data Protection Rules, 2025 — Rule 7 (breach intimation) (official text)

This guide explains the law for engineering and business teams. It is not legal advice; confirm how it applies to your organisation with qualified counsel, and check the Gazette for later notifications or amendments.

About the author

Suhas Bhairav

Suhas Bhairav

Systems architect and applied AI practitioner who designs production AI systems for Indian and global companies, with a focus on privacy-by-design under the DPDP Act and the EU AI Act.