"AI in banking must augment, not replace, human judgment." Discuss the accountability challenges posed by AI adoption in India's banking sector.

Artificial Intelligence is entering credit underwriting, fraud detection and customer service in Indian banks. RBI Governor Sanjay Malhotra, in his FIBAC 2026 address, warned that the biggest risk is not technological failure but erosion of human judgment and accountability [1]. AI must therefore remain an aid to, not a substitute for, banker discretion.

Why augmentation, not replacement

• Scale of adoption: RBI's FREE-AI Committee (2025) found AI already deployed in customer support, sales, credit underwriting and cybersecurity, with a large majority of entities exploring further use cases [2].
• Efficiency gains are real — faster fraud detection, wider credit access, cheaper grievance redressal — but they operate on decisions that materially affect citizens' financial lives.

Accountability challenges

• Diffused responsibility: banks may deflect blame to vendors or algorithms. RBI's position is categorical — "the model decided" is no answer to a customer, auditor or the regulator; responsibility rests with the bank [1].
• Opacity ("black box"): models used in loan rejection or account freezing may be unexplainable, undermining the customer's right to reasons and fair treatment.
• Bias and exclusion: training data reflecting historical lending patterns can silently disadvantage women, small borrowers or backward regions.
• Systemic risk: herding on similar models can amplify errors across institutions; RBI's Financial Stability Report flags AI in high-stakes applications as needing responsible, ethical use so outcomes do not undermine public trust [3].
• Data protection: large-scale customer data processing raises obligations under the Digital Personal Data Protection Act, 2023 [4].
• Supervisory capacity: auditing an evolving model is harder than auditing a static process.

AI is thus best seen as a decision-support layer sitting beneath, not above, human responsibility. The way forward lies in board-approved AI governance policies, model inventories, red-teaming, and "meaningful human oversight — the ability to explain, intervene and override" as a design principle [1], operationalising the FREE-AI framework's sutras of trust, fairness and accountability [2]. Technology may compute; only institutions can be answerable.