How AI Impacts Finance: Credit, Markets, Risk and Rules

Ask how AI impacts finance and most answers return a list: fraud detection, chatbots, robo-advisers. The list is accurate and it explains very little. A better starting point is the one economists at the Bank for International Settlements use. Finance, at its core, is the processing of vast amounts of information into price signals that coordinate the rest of the economy. Every leap in information processing, from double-entry bookkeeping to the spreadsheet, has changed what finance does. AI is the latest leap, and it is arriving fast: by 2025, 65 percent of financial institutions reported actively deploying AI, and sector spending is estimated at roughly 75 billion US dollars in 2026.

This piece follows that framing. It walks through what AI changes in credit, markets, insurance, payments and operations, then turns to the two places most explainers skip: the models central banks now run, and the systemic risk that appears when every institution reads the same data through similar models.

How AI impacts finance: the short answer

AI makes finance cheaper to run, faster to price and more exposed to the same few points of failure. The efficiency case is large. The World Economic Forum estimates that 32 to 39 percent of work across banking, insurance and capital markets has high potential to be fully automated, with a further 34 to 37 percent open to augmentation. Financial firms spent 35 billion dollars on AI in 2023, a figure the same report projects to reach 97 billion by 2027.

The stability case is less comfortable, and it is where the interesting questions sit. We come to it after the functions.

Credit and lending: who gets a loan, and at what price

Lending is a prediction problem: will this borrower repay? Machine learning has been applied to it for years. The BIS paper notes that ML models can reduce underwriting costs and expand access to credit for people previously excluded, although few institutions have taken advantage of the full range of these opportunities.

The practical examples are now named. S&P Global Ratings describes how Santander's corporate and investment bank built a tool called Kairos that shows how a corporate client could be affected by economic events, informing lending decisions. S&P also reports that about 40 percent of financial services companies rely primarily on machine learning for fraud detection and financial forecasting, and judges AI an incremental game changer whose effects may eventually reach its view of banks' creditworthiness.

Lenders expect more. In an earlier survey of financial institutions, the WEF found that lenders widely expect to profit from AI-enabled credit analytics, and that risk management already had the highest implementation rate of any use, at 56 percent.

Credit is also where regulation bites first. The EU AI Act sets binding, risk-based obligations with phased requirements from 2025, and treats credit, pricing and market activity as high-risk use cases that demand human oversight and transparency. A model that declines a mortgage now has to be explainable to a supervisor.

Markets and asset management: faster prices, thinner margins

Capital markets were automated long before generative AI. The IMF's October 2024 Global Financial Stability Report records that algorithmic trading now makes up about 70 percent of US equities trading and more than half of futures trading. Non-bank financial intermediaries, the funds and trading firms outside the banking perimeter, now hold over half of all financial market assets globally, and the IMF expects AI to make them more important still, because they are more agile and face fewer constraints on adoption than large banks.

What AI adds is the ability to process unstructured and high-frequency data quickly and turn it into trading decisions. S&P cites Bank of America's Glass platform, which consolidates market data across asset classes with in-house models to anticipate client needs. In capital markets, algorithmic trading is the second-largest source of AI return on investment, at 26 percent according to Statista's compilation.

Why speed cuts both ways

The IMF's view is balanced and worth reading carefully. AI-enhanced trading can deepen liquidity and make prices respond faster to new information. Thinner margins, though, can encourage more leverage across the system. And the liquidity that algorithms provide in calm markets tends to diminish under stress, when strategies are programmed to de-risk or switch off. We return to that below.

Insurance and climate risk: pricing what has not happened yet

Insurance is the third function in the BIS framework, and it is the one where AI's appetite for data meets the hardest forecasting problem in finance: risks with little history. The OECD lists fraud detection, credit decisions, risk management, compliance and portfolio management among the uses now spreading through banking and insurance, and it warns that bias, deceptive outputs and data protection failures can harm consumers along the way.

The sharpest application is climate. Physical risk to property, sovereign debt and catastrophe bonds is increasingly priced with machine learning models. We cover the models, asset class by asset class, in our piece on how AI now prices climate risk across asset classes. The short version for this article: the models have moved faster than the market's willingness to price what they find.

Payments, fraud and the back office

The least glamorous impact is the largest in headcount. Finance is language-heavy work: contracts, filings, compliance reports, client letters. That is why the automation share is so high, and why the early returns are concentrated here. The 2026 IIF and EY survey found that institutions are already realising return on investment from predictive AI (86 percent) and generative AI (81 percent), with cost reduction the most significant benefit. Cost is also the obstacle: 50 percent named the expense of implementing AI as a key challenge.

Access inside firms has widened quickly. Deloitte reports that surveyed financial services companies doubled worker access to sanctioned AI tools in one year, from 30 to 62 percent of staff.

Customers are moving too, with a condition attached. In the WEF's 2026 playbook, 71 percent of banking consumers said they would welcome an AI assistant in their bank's app, yet 82 percent want to approve an agent's actions before they are executed. That second number defines the next design problem in payments, which our colleagues at EX Fi examine in detail in how an AI agent gets permission to spend.

Macro modelling: AI inside the central bank

Most explainers stop at the firm. The public side of finance is changing as well. The Financial Stability Board noted as early as 2017 that central banks expected to use big data for economic forecasting, particularly for inflation and prices, and that 39 percent expected to nowcast home prices with it. Natural language processing applied to news and filings can help authorities detect and anticipate market volatility, liquidity risk and financial stress.

The gap is capacity. The 2026 Cambridge Centre for Alternative Finance report surveyed 130 regulatory authorities and found that 48 percent were still exploring AI or not engaged with it at all. Among industry respondents, 40 percent reported advanced adoption, double the 20 percent of regulators. The institutions doing the supervising are, on average, a step behind the institutions being supervised.

This is the question that macro-finance researchers now work on: how to model an economy whose financial sector reacts at machine speed. It is the territory of researchers such as Willi Semmler, Henry Arnhold Professor of International Cooperation and Development at The New School for Social Research and Senior Researcher at IIASA, who is among the confirmed speakers at EX Future Summit.

The systemic risk: when every model reads the same data

Here is the part of how AI impacts finance that matters most to policymakers. In 2024 the FSB identified four AI-related vulnerabilities that could raise systemic risk: dependence on a concentrated set of third-party providers, rising market correlations, cyber risk, and model risk from limited explainability and uncertain data quality.

The first two compound each other. Accelerated computing chips and cloud services are dominated by a small number of firms, and many large language models share architectures and are trained in part on the same web data. Models built on the same foundations tend to make correlated predictions. The FSB's sharpest line is that a segment of the financial sector using the same models and data could be hit by a shock as if it were a single institution. It also flags the risk of automated systems pulling their kill switches at the same moment.

The IMF describes the mechanism in markets. In normal times, AI may uncover diverse strategies that make markets more resilient. Under an adverse shock, models could simultaneously rebalance toward safe assets, creating a self-fulfilling spiral of fire sales. And if capital markets come to rely on a handful of AI providers, the failure of one could produce stress comparable to the failure of a clearing house. The IMF's recommendations are concrete: re-parameterise circuit breakers for AI-driven trading, review margin requirements for rapid price moves, and bring critical AI service providers into supervisory view.

Adoption is ahead of governance

The numbers line up into one picture. Cambridge finds 81 percent of financial firms adopting AI at some level and agentic AI already in active adoption at 52 percent, yet only 14 percent see AI as transformational to their strategy. Deloitte finds 71 percent of firms plan to deploy agentic AI within two years, while 23 percent report a mature governance model for autonomous agents. The IIF and EY survey is more reassuring on paper: every institution with AI in production governs it through formal risk processes, and 56 percent have built new frameworks specifically for it.

The direction is clear. AI is moving finance from analysts who process information to systems that act on it, and the rules for that shift are being written now, mostly by supervisors who are still building their own capacity. Statista's figure that 40 percent of institutions saw AI raise profitability by early 2026, while 43 percent saw no change, suggests the payoff is real but uneven.

That combination, fast industry, slower regulators and researchers modelling the gap, is why the Finance and Government tracks sit side by side in the nine-track program of EX Future Summit, running 18 to 20 November 2026 online, in Las Palmas and in Bali. Online attendance is free for verified researchers and students. For how this summit compares with others on the calendar, see our guide to the AI in finance summit landscape, and for the broader overview of the field, our primer on AI in finance.

FAQ

Will AI replace financial analysts?

Mostly it will change what they do. The WEF estimate puts 32 to 39 percent of financial services work at high potential for full automation, but a similar share, 34 to 37 percent, is better suited to augmentation, where the analyst stays and the routine processing goes.

Is AI making financial markets more volatile?

Not in normal conditions, where it can deepen liquidity. The IMF's concern is stress: correlated models reacting the same way, and algorithms that withdraw liquidity or shut down when prices move in ways they have not seen before.

How is AI in finance regulated?

In the EU, the AI Act treats credit, pricing and market activity as high-risk uses with phased obligations from 2025. Elsewhere, supervisors mostly apply existing model risk and third-party rules, and the FSB and IMF both urge closer oversight of critical AI providers.

Is AI actually profitable for banks yet?

For some. By early 2026, 40 percent of financial institutions reported that AI had increased overall profitability and 43 percent reported no change, per Statista's compilation.