Reserve Bank of India proposes framework for AI adoption in India’s finance sector

Principles and priorities

The committee presented a series of proposals to make AI more accessible and accountable. Central to this is a financial data infrastructure for the whole sector, connected to India’s AI datasets platform (AIKosh). This would give banks and financial technology firms standardised datasets for training AI models. The committee also suggested a framework to connect AI with digital public platforms such as the Unified Payments Interface (UPI). 

To encourage safe experimentation, it proposed an AI innovation sandbox for companies to test algorithms using shared computing power and selected data before real-world use. It also called for shared graphics processing unit (GPU) “landing zones” that smaller lenders can rent, ensuring AI isn’t monopolised by big banks.

To build a more democratic AI ecosystem, the committee recommended creating “digital public intelligence” and developing local AI models trained on Indian regulations, financial products and languages. The goal is to establish a system of voice-enabled banking, multilingual services and built-in fraud checks that can improve financial access for millions.

Experts warned, however, that gaps in data infrastructure and talent may slow adoption. Ashish Kakar, director of financial insights at IDC Asia-Pacific, noted that the biggest challenge for Indian banks is data spread across distributed systems and lengthy extract, transform, load (ETL) processes. He also noted that while tech talent is available, a lack of banking knowledge can result in wasted investment.

The committee highlighted AI’s potential for financial inclusion and suggested using alternative data, such as utility payments and Goods and Services Tax (GST) filings, to assess the creditworthiness of borrowers who are new to credit. Kakar said this focus on inclusion will drive growth and help Indian banks become global leaders.

Guardrails and governance

To ensure accountability, the committee recommended that boards approve AI policies and set up governance structures within financial institutions and a standing committee to monitor new risks, which range from biased lending models and deepfake fraud to the systemic weakness that could arise if too many firms use the same algorithms.

While the RBI’s principle-based, consultative approach offers a strong base for AI oversight, Aruna Pannala, partner at Deloitte India, noted that model risk management guidelines are yet to be formalised. On data privacy, Pannala said the committee’s stance is broadly aligned with the Digital Personal Data Protection (DPDP) Act, but still needs minimum standards to avoid inconsistent interpretations and customer grievances.

Shreya Suri, partner at CMS Induslaw, highlighted differences in the threshold for reporting data breaches. While the DPDP Act mandates disclosure of all personal data breaches to boards and individuals, the proposed AI framework suggests a tiered system based on severity. She said this reflects an emphasis on anonymisation and privacy, even though datasets may still carry personal information.

How these overlapping rules are resolved will ultimately depend on the final framework, as the current proposals remain recommendations.