On August 13, 2025, the RBI released its FREE-AI Framework with 26 recommendations for responsible AI in finance. While ambitious, it overlooks some ground realities of the Indian financial sector. First, it lacks enforceable mechanisms to prevent AI systems from excluding the 90% of Indians in the informal economy whose irregular incomes make them invisible to algorithms trained on urban salaried data. Second, it imposes identical governance requirements - AI Centres of Excellence, mandatory audits, board policies, on both large banks and small cooperatives, creating compliance costs that trigger consolidation and reduce rural financial access. Third, it ignores how synchronized AI decision-making across institutions creates systemic risk: when multiple banks use similar algorithms that flag the same sectors during downturns, coordinated credit contraction amplifies recessions instead of stabilizing them. The framework needs structural compliance, measurable inclusion standards, and counter-cyclical safeguards.
On August 13, 2025, the Reserve Bank of India released its Framework for Responsible and Ethical Enablement of Artificial Intelligence (FREE AI), a landmark initiative outlining how India’s financial sector should adopt AI responsibly. The framework lays out 26 recommendations aimed at making AI adoption in finance safe, orderly, and inclusive. It promises innovations like AI-powered customer service, better fraud detection, and even offline solutions for digital payments.
However, despite its ambition and breadth, the FREE AI overlooks key ground realities of India’s financial landscape. The framework's one-size-fits-all approach, though well-intentioned, is deeply out of touch with how financial services function across India's diverse institutional landscape and population segments.
First, the FREE AI provides no enforceable mechanisms to prevent AI systems from excluding marginalized populations. The FREE AI aspires to make AI “accessible for everyone,” but beyond broad goals like offline access and inclusion, it lacks a concrete plan to reach India’s financial margins. While the Reserve Bank of India cannot bridge the broader digital divide, its framework shapes how financial institutions design and deploy AI. The core problem lies in how AI systems are trained: if credit algorithms rely predominantly on urban data from salaried borrowers with formal income documentation, they will disadvantage agricultural workers, informal entrepreneurs, and low-income households whose earnings are irregular or undocumented. In India, where approximately 90% of the workforce operates in the informal sector (ILO, 2018), this bias affects many potential borrowers. Studies show that machine learning models trained mostly on formal-sector data often misjudge creditworthiness when applied to informal or low-income borrowers, even when these workers have strong repayment capacity (Berg et al., 2020). This could make access to credit increasingly conditional on digital visibility and formal credentials.
Second, it imposes disproportionate compliance burdens on smaller institutions. The FREE AI mandates that financial institutions develop in-house expertise and establish AI Centres of Excellence. It also requires every institution to adopt a board-approved AI Governance Policy along with mandatory audits, and consumer disclosure requirements. This may sound reasonable for larger banks but ignores the reality of smaller financial institutions. Smaller institutions like cooperative banks typically deploy rudimentary automated tools like simple customer service chatbots or loan assessment algorithms. Yet, under the FREE AI, they face the same governance requirements as those implementing sophisticated machine learning systems across a much higher volume of transactions. This creates a regulatory scale mismatch; rules designed for large institutions being applied uniformly to small ones, creating disproportionate costs. Research on financial sector regulation demonstrates that uniform compliance requirements systematically disadvantage smaller institutions, often triggering consolidation as the economically rational response (RBI, 2024). This reduces competition and could leave rural and semi-urban areas with fewer local financial institutions that understand their specific needs.
Third, it ignores systemic risks from synchronized algorithmic decision-making. Specifically, the FREE AI focuses heavily on individual institution governance but misses how AI systems, when trained on similar datasets and using similar techniques could create unintended systemic effects. Human bankers have different perspectives and risk appetites. During economic downturns, this diversity acts as a stabilizer; some banks tighten lending while others continue, some retreat from specific sectors while others see opportunity. This staggered response provides crucial liquidity and prevents sharp credit crunches (Acharya & Mora, 2013). However, AI models that use industry-standard training approaches and similar macroeconomic datasets, are more likely to generate correlated risk assessments. If many financial institutions use similar lending algorithms that flag the same sectors as high-risk during an economic slowdown, they may all cut credit to those sectors at once. This synchronized reaction can worsen economic downturns instead of easing them - a pattern known as procyclicality (Danielsson et al., 2016).
India needs a fundamental overhaul from technology-first to people-first AI implementation in financial services. To address accessibility gaps, the framework should mandate staged rollouts starting with non-critical urban applications before expanding to rural credit decisions. Institutions must be required to use regional training datasets that capture local business patterns. Creating scale-appropriate regulation through a three-tier compliance system where small cooperative banks face streamlined governance requirements via shared oversight cooperatives, medium institutions get extended implementation timelines with government technical support, and only large banks face immediate full compliance. Further, financial institutions must embed economic resilience into their AI systems through counter-cyclical programming trained on recession data. The framework should mandate diversity protocols that prevent synchronized credit contraction.
The real test isn't whether AI can detect fraud or speed up transactions, it's whether a farmer in a remote village and a software engineer in a metro city have equal access to financial opportunity in an AI-driven system. The technology is already here- what will define its impact is not its sophistication, but how responsibly it is designed and deployed. The FREE AI captures the right governance vision, but without enforceable accountability mechanisms, its implementation risks deepening the very inequalities it aims to bridge.
REFERENCES
Acharya, V. V., & Mora, N. (2013). A crisis of banks as liquidity providers. Journal of Finance, 70(1), 1-43.
Berg, T., Burg, V., Gombović, A., & Puri, M. (2020). On the rise of FinTechs: Credit scoring using digital footprints. Review of Financial Studies, 33(7), 2845-2897.
Danielsson, J., Valenzuela, M., & Zer, I. (2016). Learning from history: Volatility and financial crises. Review of Financial Studies, 31(7), 2774-2805.
International Labour Organization (ILO). (2018). Women and men in the informal economy: A statistical picture (3rd ed.). Geneva: ILO.
Reserve Bank of India (RBI). (2024). Report on Trend and Progress of Banking in India 2023-24. Mumbai: Reserve Bank of India.