"The model decided" cannot be an acceptable regulatory defence. Analyse the ethical and legal implications of algorithmic decision-making in critical public-facing sectors.

Algorithmic systems now decide who gets a loan, a benefit or a flagged transaction. RBI Governor Sanjay Malhotra's FIBAC 2026 caution — that "the model decided" can never answer a customer, auditor or regulator — captures the core problem: automation diffuses responsibility even as it concentrates power. Decomposed below, the challenge is ethical in origin and legal in remedy.

Ethical dimension

• Opacity vs. informed consent: black-box scoring denies the citizen reasons for an adverse outcome; RBI's FREE-AI framework therefore places explainability among its guiding sutras [1].
• Bias and exclusion: models trained on past lending or policing data reproduce historical disadvantage against marginal groups.
• Moral deskilling: routine deference to model output erodes human judgment, the very risk Malhotra flagged in banking [2].
• Diffused accountability: bank, vendor and developer each point to the other — a "problem of many hands".

Legal dimension

• Fairness under Article 14: arbitrary, unreasoned automated action in State or State-adjacent functions is vulnerable to review; reasoned orders are a natural-justice minimum.
• Data and privacy duties: the DPDP Act, 2023 imposes purpose limitation, accuracy and grievance-redressal obligations on data fiduciaries deploying such systems [3].
• Sectoral regulation: RBI's Digital Lending Directions, 2025 mandate Key Fact Statements and named grievance officers — locating liability on the regulated entity, not the app [4].
• Liability gap: no statute yet defines fault for autonomous model error; outsourcing decisions cannot outsource duty.

Way forward

• Board-approved AI governance policies with model inventories and audit trails [1].
• "Human-in-the-loop" for rights-affecting decisions — power to explain, intervene and override.
• Right to reasons and appeal to a human authority; periodic bias and impact audits.

Algorithms should widen access to credit, welfare and services, not narrow the citizen's remedy. India's approach — enable adoption while fixing accountability on the deploying institution — is the balanced path, provided oversight is designed in rather than bolted on. Technology may compute the decision; the institution must always own it.