.png)
Dr. Mishra is former Executive Director of RBI and the Founder Director of its College of Supervisors. He is currently RBI Chair Professor at Gokhale Institute of Politics and Economics.
August 24, 2026 at 5:00 AM IST
In 1950, Alan Turing framed the debate over artificial intelligence with a simple question: "Can machines think?" Ray Kurzweil later popularised the idea of a technological singularity, the convergence of human and machine intelligence.
For finance, the question is no longer abstract. Banks use AI to detect fraud, assess credit, monitor transactions, and support decisions. The trajectory is moving from process automation towards market transformation and, eventually, more autonomous systems.
That makes AI a double-edged sword. It can improve risk identification and management, but it can also create risks that are faster, more correlated, and harder to see. Regulation must ensure that adoption strengthens institutions without weakening the financial system.
Systemic Faultlines
Explainability, accountability, and ethical deployment are therefore essential. An institution must be able to show how a model reached a decision, who is responsible, and whether the outcome complies with law and policy. The risk becomes systemic when firms rely on similar data, vendors, or model architectures.
Algorithmic clustering is the central faultline. AI can accelerate market reactions, strengthen feedback loops, and produce synchronised responses. Under stress, common signals can prompt common trades, credit decisions, or simultaneous cuts in risk appetite, amplifying herd behaviour, liquidity shocks, leverage sensitivity, and flash events.
The danger extends beyond banks. As AI widens non-bank participation and raises productivity, it can deepen interconnectedness. A model that performs well at one firm may still worsen system-wide outcomes when many firms use it simultaneously. It must therefore be judged by both firm-level accuracy and the collective behaviour it induces.
Charles I. Jones captured the broader trade-off in his 2023 NBER paper, "The A.I. Dilemma: Growth versus Existential Risk". AI can accelerate growth, but poorly aligned systems can generate risks that expand with their capabilities. In finance, a tool that makes every institution faster can make the system less stable if all react alike.
Governance Agenda
Use case requires an institution to define what a system does instead of applying generic controls to every deployment. A customer-service chatbot, a fraud-detection engine, and a credit-underwriting model do not pose the same risks.
Materiality determines the intensity of governance. The greater an application's potential effect on operations, customers, legal obligations, or regulatory compliance, the stronger its controls, auditability, and human oversight should be.
Proportionality links governance to institutional size, complexity of use, and potential harm. It does not reserve AI for large firms. Smaller institutions may benefit substantially, but their controls must match their capabilities and exposures.
Sequencing matters too. Denis Beau, First Deputy Governor of the Bank of France, argued in September 2025 that institutions should simplify inefficient processes before automating them. AI should not preserve unnecessary complexity. Automating a flawed process makes it faster, not safer.
The Reserve Bank of India's FREE-AI framework places fairness, responsibility, explainability, and ethics at the centre of adoption. It treats governance as more than compliance, calling for structured risk governance, AI inventories, audit protocols, cross-functional oversight, inclusion, and trust. Its graded liability approach also seeks to prevent genuine first-time errors from deterring responsible experimentation.
Regulators face a parallel challenge. AI can improve fraud detection, transaction monitoring, and forward-looking supervision by identifying weak signals and atypical behaviour. But adoption at scale makes supervision harder. Regulators will need their own tools to understand models, map networks, monitor exposures, and spot concentrations.
The hardest risks may remain invisible until institutions react together. Algorithmic sameness can produce synchronised trading, uniform credit denials, or common mispricing. Addressing it requires model diversity, cross-institutional stress testing, vendor transparency, and human capacity to challenge model outputs.
For boards and senior management in India, five priorities follow.
1. Calibrate risk appetite dynamically. A risk appetite framework must reflect institutional conditions and the economic cycle. AI can translate board-approved limits into operating thresholds and flag exposures moving outside them.
2. Detect faultlines early. Use AI to identify vulnerabilities, weak controls, and emerging concentrations before they become losses. The objective is preventive risk management, not faster reporting after an event.
3. Measure interconnected risks. Credit, market, liquidity, and operational risks increasingly interact. AI-based econometric tools can separate these effects and show how a shock in one category migrates into another.
4. Establish board-level governance. Every institution needs a board-approved responsible AI policy and an interdisciplinary committee spanning risk, technology, legal, compliance, and business functions. It should monitor model drift, bias, and performance degradation, and oversee stress tests.
5. Tighten third-party controls. Vendor audits should cover model lineage, data sources, audit trails, validation standards, and contractual access to information. Black-box arrangements are incompatible with accountability. Material third-party models should be reviewed at least quarterly, and more often where risk warrants.
The objective is not to slow AI in finance, but to keep speed from outrunning resilience. AI can make institutions more efficient and supervision more intelligent only if governance keeps pace with capability. Otherwise, tools that manage risk at the firm level may create risk for the system.
More From BasisPoint
