AI in banking: Artificial intelligence is already part of finance. Banks use it to assess borrowers, detect fraud and handle customers. Fund managers use machine learning to process market information and trade. Insurers deploy it in underwriting and claims. The gains are substantial: lower costs, faster decisions and better use of data.
The same speed can turn a local failure into a market-wide event. The IMF’s latest assessment warns that AI could make financial markets more tightly coupled, deepen reliance on a few technology providers and allow cyberattacks to spread faster. These are familiar financial risks, but compressed into much shorter periods.
READ | Thirty-five years after, economic reforms confront a new constraint
AI adoption in Indian banking
Indian banks use AI in credit assessment, customer service and fraud detection. The RBI’s regulatory sandbox provides a controlled setting for financial firms to test new products. In August 2025, the central bank also released its FREE-AI framework, which sets out principles for the responsible use of AI in finance.
The RBI Innovation Hub’s MuleHunter.AI shows what the technology can do. The model analyses transaction patterns to identify mule accounts used to move the proceeds of cybercrime. Banks have traditionally relied on fixed rules and manual investigation, which struggle to keep pace with the number and variety of suspicious transactions.
AI can also widen the information available to lenders. GST returns, bank statements and digital payment records may help assess small businesses and first-time borrowers who lack conventional credit histories. This could improve lending decisions, though more data do not necessarily eliminate bias or poor underwriting. They may merely give these faults a more elaborate statistical form.
A borrower denied credit by an algorithm must be told why. He must also be able to challenge an incorrect GST entry, payment record or bank transaction used by the model. The lender cannot pass responsibility to the technology vendor. A bank that uses the model must answer for its decision.
READ | AI needs a knowledge dividend for the human work it uses
AI in banking and financial market volatility
The larger risk lies in markets. Trading systems increasingly process the same economic releases, company filings and price signals. If several models read a shock in much the same way, they may buy or sell together.
Herding is hardly new. AI changes its speed. A fall that once unfolded over hours could be compressed into minutes as models rebalance portfolios, lenders demand more collateral and leveraged investors sell assets. Liquidity can disappear precisely when it is most needed.
The IMF says future flash crashes may arise from several AI systems responding in parallel to the same information, rather than from a coding error. Its 2024 Global Financial Stability Report found that AI-based trading could raise market speed and volatility under stress, especially when strategies become correlated. It asked regulators to review circuit breakers, margin requirements and the resilience of market infrastructure.
This remains a prospective risk. The IMF found that most current use of AI in capital markets builds on existing machine-learning and quantitative methods. Fully autonomous trading is still limited. The concern is that wider use of similar models could make markets more correlated and leave regulators less time to act during a sell-off.
Regulators will need information that they do not routinely collect. They must know which institutions use AI in trading, which models or data sources they share and how these systems behave under stress. Conventional capital and liquidity rules remain necessary, but they were not designed to trace common dependencies among algorithms.
READ | Standard Chartered India trims retail for affluent clients
Cloud concentration becomes a financial risk
Large AI systems require computing capacity, cloud storage, specialised chips and foundation models. Most banks cannot build this infrastructure themselves. They buy it from a small group of technology companies.
A failure at one provider could then affect several banks, insurers, payment firms and market intermediaries at once. Financial regulators have long supervised institutions whose collapse could damage the system. They must now map technology companies that have acquired systemic importance without becoming financial institutions.
This dependency also weakens oversight. A bank remains responsible for a model supplied by an outside vendor, but its own risk officers may not understand the model or have access to the information needed to test it. Contractual outsourcing does not transfer regulatory liability.
AI raises the cost of cyber failure
AI helps banks identify unusual transactions and network activity. It also makes phishing, malware creation and automated fraud cheaper. Attackers can search for vulnerabilities, alter their methods and target many institutions at once.
The IMF estimates that severe cyber incidents can cause funding strains, solvency problems and wider market disruption. Shared software, cloud services and payment networks increase the chance that a single weakness will affect several institutions. IMF analysis of AI-enabled cyber risk
This places cyber resilience within the remit of financial stability. The RBI, SEBI and other regulators cannot leave it entirely to the technology departments of regulated firms. Stress tests must cover prolonged cloud outages, compromised models and attacks on services used by several institutions.
Financial regulators need AI expertise
Supervisors cannot scrutinise models they do not understand. They need specialists capable of examining training data, assumptions, biases and failure conditions. Shortages of such skills are more acute in emerging economies, where regulators compete with banks and technology companies for the same people.
Consultants can fill some gaps, but dependence on vendors creates another conflict. A regulator that relies on the firms supplying the technology may find it difficult to challenge their models or demand disclosure.
AI can help supervisors scan transactions, detect market abuse and identify unusual concentrations of risk. It cannot carry regulatory responsibility. Decisions affecting credit access, market integrity or financial stability must remain attributable to named officials and institutions.
India has good reason to encourage AI in finance. Better fraud detection and credit assessment can lower costs and extend formal finance. But adoption has moved ahead of regulators’ knowledge of where models are used, which outside providers support them and how several systems might react to the same shock. The next financial accident may give the authorities only minutes, rather than days, to find out.