Operationalizing RBI’s Free AI Framework and Model Risk Management in the Indian BFSI Sector.

Introduction and Background: 

The Indian BFSI sector is going through a change because of  digitalization and this change is being driven by the use of artificial  intelligence and machine learning. Smart computer programs are  being used more and more for tasks like checking how likely  someone is to default on a loan making trades using computers  finding fraud helping customers with their questions and following  rules about money laundering. These programs make things faster  and more efficient.

They also come with some dangers. The  dangers include things like decisions made by machines not  knowing how some programs work mistakes made by AI that  creates content problems with keeping personal information private  attacks on computer systems and relying too much on one  company for these tools. To deal with these dangers and to create  a set of rules the Reserve Bank of India (RBI) has created two new  tools. First on August 13, 2025 the RBI shared a report from a  group of experts that talked about how to use intelligence in a  responsible and fair way.

This report helps build trust and makes  sure that AI is used in a way that’s fair and accountable. Second on  June 24, 2026 the RBI shared a draft document that talks about  how to manage the risks of using models. This document includes  all intelligence and machine learning models, not just the ones used  for credit scores. It covers the time from when a model is made to  when it is used. The goal is to see how companies, in India should  use the rules from the report and the new guidelines. Making sure  AI is used in a way and managing the risks it brings is now not just  something companies are encouraged to do but it is something they  must do.

Objectives: 

  • To conduct an extensive research of the principles, strategic  foundations and recommendations stated in the RBI’s Free AI Framework. 
  • To analyze the requirements in RBI’s Draft Guidance on Model  Risk Management issued in June 2026 with regard to governance,  risk tiering and validation independent of the model developer.
  • To identify the challenges Indian financial institutions might face  while shifting from model validation to AI governance.
  • To design a strategy for implementation of the technology  innovation in accordance with regulatory requirements.

Analysis:  

AI Governance in the Indian BFSI Sector entails an elaborate  comprehension of the interaction between the ethical principles and  the risk management framework, which are quantified. The  regulatory framework of RBI utilizes two means to govern AI in  BFSI sector: the ethical “Sutras” of Free AI and operational  guidelines of Model Risk Management Framework.

The Free AI Framework: Principles and Pillars: 

The Free AI Framework is founded on seven core principles known  as “Sutras”. They are as follows: 

Trust is the Foundation: AI systems should be built to instill and  sustain trust in the financial system’s integrity. 

People First: Human decision-makers should have ultimate  decision-making power in order to ensure that human interests are  at stake when using automated systems. 

Innovation over Restraint: The regulation encourages responsible  innovation and the development of safe cases, not prohibition. 

Fairness and Equity: The models should not show any kind of  unlawful bias and discriminatory behaviour towards certain  individuals or groups of people. 

Accountability: Regulated institutions remain fully accountable for  the decisions and actions of their artificial intelligence (AI) systems,  regardless of whether they are developed by the institution itself or  by third parties outside the institution. 

Understandable by Design: AI system should have explainable  features like documentations to track and understand the  automated logic by all users. 

Safety, Resilience and Sustainable: The systems shall be  resilient and robust against cyber-physical vulnerabilities and  infrastructure disruptions. 

The following is how the RBI translates the above seven principles  to actionable steps in the form of six strategic pillars: Infrastructure,  Policy, Capacity, Governance, Protection, and Assurance. The six  strategic pillars contain 26 actionable recommendations altogether.  

In terms of innovation and infrastructural recommendations, for  example, the RBI proposes to set up common compute  environments, AI innovation sandboxes, and sector-specific large  language/credit models to give equal access to smaller financial  entities. When it comes to regulatory and compliance issues, it insists on adopting a proportionate risk-based approach and  applying strict criteria to those applications which have significant  risks while being flexible towards low-risk utilities. Model Risk  Management Guidance (June 2026) While the Free AI Framework  deals more with ethical guidelines, the 2026 Draft Guidance on  Model Risk Management contains the operational requirements.  The practice of MRM was limited to legacy credit underwriting  scorecards before 2026. Yet, the new 2026 guidelines extend this  concept by defining model risk management as managing risks  from any quantitative method/system/algorithms which input  information into processing to come up with statistical  estimates/classifications/recommendations. It directly regulates  advanced machine learning models, deep learning networks, and  generative AI platforms. 

The MRM guidance outlines an elaborate model lifecycle  governance expectation, composed of several important  components. They are as follows: 

Model Inventory and Version Control: All regulated firms must  create and maintain an inventory system that dynamically tracks all  active, inactive, under development and retired models. None of the  models can become operational without being formally logged in  this register.  

Risk-Tiering System: All models developed by the financial  institutions need to be classified into low, medium and high-risk  categories. Contrary to the traditional approach that relied on  materiality of money alone, in the 2026 guidelines, it was clarified  that the criteria used to determine risk-tiering must include three  important vectors, namely, complexity of the technology, entity’s  dependence on the model outputs and degree of autonomy given to  the model in taking its own decisions in the absence of human  intervention. 

Model Validation: An important component of the architecture is  the independent validation of all credit and decision-making AI  models done by an independent team or a group, which is not  associated with the model development process. Prior to model  deployment, this independent team evaluates the model for  conceptual validity, stability, data lineage and software  vulnerabilities.  

Human Involvement and Controls Over Model Interventions: Human involvement and control over model interventions are  mandatory according to the guidance through well-defined  architectures such as “human-in-the-loop” (which requires human  review and approval of every automated output) or “human-on-the loop” (where automated systems run on their own, but there are  realtime override and kill-switch capabilities to prevent any  uncontrolled activity). 

Convergence Issues in Indian BFSI: 

BFSI Banks and NBFCs in India face unique technical and practical  problems in converging Free AI standards and the MRM guidelines.  The first issue that arises from this convergence is the problem of  the “black-box” nature of the deep learning algorithms and  ensembles used. Traditionally, validation was done using  coefficients of linearity, but the new AI techniques work through  multi-dimensional, non-linear matrices and thus are hard to make  explainable. When it comes to the convergent framework, if the  logic behind the model cannot be explained and related to human  logic, then the principle of “understandable by design” would not be  satisfied and would not meet the MRM documentation  requirements. The second operational issue is the quality of data,  concept drift, and data privacy. The quality of training input data is  very crucial when using AI models. Within the Indian credit  scenario, using unstructured data like utility bills, social footprints, or  local smartphone data might lead to the formation of proxy variables  leading to discrimination against marginalised demographics.  Models experience “concept drift”, meaning macroeconomic 

changes or black swan market conditions cause the real-world data  patterns to differ from the patterns from which it was trained,  reducing its accuracy and unmonitored financial risk. Finally, the  third-party dependency  

vector leads to compliance risks. Almost all the Fintech firms, mid sized NBFCs, and cooperative banks do not build their AI models  from scratch but rely on third-party tech vendors for specialised  models and API services. According to the RBI’s MRM guidelines,  relying on AI models provided by third-party vendors does not serve  as an excuse for operational failures or non-compliance. When the  vendor declines to reveal proprietary training data and weights,  citing intellectual property concerns, the regulatory entity becomes  responsible for the software’s biases, hallucinations, and  anomalies. 

Findings and Recommendations: 

Following a detailed review of regulatory documentation and sector  implementations, the following findings have emerged:  

AI risk has evolved far beyond the mere credit loss issue to cover  areas of ethical fairness, technical transparency, and algorithmic  security, making previous IT governance procedures inadequate. 

Ultimate accountability for the governance of AI has clearly been  shifted to the top; in this respect, the RBI regards the issue as non delegable at the board level. 

Vendor models represent a compliance blind spot because of the  lack of access to data lineages and intellectual property structures. 

In order to effectively manage the above-mentioned changes,  financial institutions operating in India must implement the following  strategic roadmaps: 

Adopt Board-Approved AI Governance Policy and Committee: 

Each regulated institution must urgently devise and implement  enterprise-level, board approved AI policy. The process of 

responsibility assignment will take place through creation of a  dedicated multi-party Responsible AI Committee which will involve  representatives from risk management, compliance, data  engineering, product development and legal departments. The above-mentioned committee will be responsible for establishing the  risk appetite for the institution and will validate all deployments  classified as high-risk by the bank. 

Develop Dual-Vector Risk Tiers Matrix: 

Financial institutions will need to build their internal catalogues for  classification of all uses cases for Ai implementation. Each use  case must be assigned the proper risk classification (low, medium,  high) based on intersection of materiality and operational  dependency. Materiality includes financial and customer impact.  Operational dependency involves model complexity, degree of  automation and lack of human oversight. High-risk deployments like  retail loans approval engines and algorithmic collections need to go  through validation. 

Deploy Advanced XAI and Fairness Toolkits: 

Tools that satisfy the transparency requirement include integrating  mathematical explainability measures into the development  pipeline. Deploying algorithms such as SHAP (Shapley Additive  Explanations) or LIME (Local Interpretable Model-agnostic  Explanations) would allow developers to decompose complicated  non-linear models into human-interpretable feature importance  breakdowns. Moreover, fairness testing for disparate impact and  demographic parity in the model must be implemented as part of  the automatic validation pipeline to avoid algorithmic discrimination  due to proxy variables. 

Mandate Realtime Red-Team Testing, Drift Monitoring, and  Logging AI Incidents: 

The evaluation of model risk is not a one-time check. The RE must  establish realtime monitoring systems to detect input data drift and  change in output distribution. Alerting systems should trigger fall back procedures or manual overrides if there is a fall in model 

performance under certain statistical benchmarks. Additionally,  regular red-team simulations must be conducted as part of the  validation to test the AI model’s robustness against malicious  prompt injection and cyberattacks. It can also help to monitor model  performance during tail risks in the economy. There should be a  logbook for tracking AI incidents and continuously improving  institution’s defenses. 

Overhaul Third-Party AI Systems Service Level Agreements  and Vendor Diligence: 

There needs to be a framework for procurement of third-party AI  systems. There should be contractual provisions obligating vendors  to provide comprehensive model documentation and data lineage,  and audit access to models. In case a vendor uses an auditable  proprietary AI system, the RE should implement wrapper methods,  benchmarking testing, and human-in-the-loop validation procedures  to manage these risks. 

Conclusion: 

Regulatory mandates of the Reserve Bank of India including the  August 2025 Free AI report and the June 2026 Draft Guidance on  Model Risk Management indicate the dawn of a new era where  artificial intelligence shall be regulated in the banking and finance  industry. In addition to mandating banks and NBFCs to have  policies that regulate the use of AI, this set of guidelines has moved  the governance of artificial intelligence from the scope of optional IT  innovation to the scope of regulatory compliance and risk  management at the board level. The execution of such regulations  calls for a delicate balance where although the regulated entity  should have strong mechanisms in place like independent  validation teams, dynamic risk-tiering and explainability tools, they  should not end up hampering the revolutionary benefits of artificial  intelligence towards financial inclusion and efficiency of processing  across India.

References: 

  • The Press Release of RBI (2024): Constitution of the Committee on  Framework for Responsible and Ethical Enablement (Free) of  Artificial Intelligence (AI) in the Financial Sector. Central Office,  Mumbai. 
  • RBI (2025): Report of the Committee on Framework for  Responsible and Ethical Enablement of Artificial Intelligence (Free  AI Framework). Department of Fintech, Mumbai. 
  • RBI (2026): Draft Guidance on Regulatory Principles for Model Risk  Management. Department of Banking Supervision / Regulation,  Mumbai. 
  • Risk Management Association of India (RMAI). (2026). Special Insights on AI Adoption and Model Risk Vulnerabilities in Indian  Commercial Banking. Portal Documentation.

Authored by:

Name: Aditya Dandavate

Course: AI and Emerging Technologies in Banking 

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.