Algorithmic decision-making in public sector banking raises concerns of accountability and bias. Critically analyse in the context of AI-based loan underwriting.

Algorithmic underwriting uses machine-learning models on alternative data — GST returns, bank statements, credit-bureau scores — to sanction loans with minimal human discretion. With SBI underwriting nearly ₹1 lakh crore of MSME loans (up to ₹5 crore each) through its Business Rule Engine in FY26 [1], the practice is now systemically significant, making its gains real but its accountability deficits urgent.

Merits: efficiency, inclusion and objectivity

• Bridging the credit gap: RBI's Expert Committee (2019) estimated an MSME credit gap of ₹20–25 lakh crore [2]; data-based underwriting shrinks turnaround from weeks to minutes.
• Cash-flow lending over collateral: GSTN and ITR-linked assessment lets thin-file, new-to-bank enterprises borrow without property collateral [1], complementing CGTMSE guarantee cover [2].
• Consistency: rule-driven scoring curbs subjectivity and rent-seeking in branch-level discretion, and frees relationship managers for higher-value work [1].

Concerns: accountability and bias

• Opacity: complex models yield no reasoned order for rejection, weakening natural justice and the borrower's right to appeal.
• Data-driven exclusion: reliance on GST and digital footprints structurally disadvantages informal micro-units — a large share of the 7.9 crore Udyam/Udyam Assist registrants are informal micro enterprises [2] — risking proxy discrimination by region, gender or sector.
• Diffused liability: outsourcing to fintech lending service providers blurs responsibility, which the RBI (Digital Lending) Directions, 2025 address through consent, audit trails and mandatory credit-bureau reporting [3].
• Systemic and privacy risk: herd-like model behaviour can amplify procyclicality; pooling GST, ITR and account data engages data-minimisation obligations.

Efficiency and equity are therefore not opposing goals but require a governance scaffold. The RBI's FREE-AI Committee (2025), with its seven "sutras" and six pillars — including proportionate, risk-based regulation, human oversight and assurance — offers exactly that template [4]. Adopting explainability standards, periodic bias audits and a human-in-the-loop for adverse decisions would let AI advance Article 38's mandate of distributive justice, turning algorithms into instruments of inclusion rather than silent gatekeepers.