The last global banking crisis was driven by mortgages, leverage and institutions that all believed the same thing at the same time. The next one will be driven by machines doing exactly the same thing, except they will do it in milliseconds.
For the last few years, everyone has talked about artificial intelligence (AI) as though it will be the greatest productivity tool ever invented. Specifically, in banking, for fraud and compliance purposes, AI is seen as a saviour. AI will write the code, answer the customer, detect the fraud, assess the credit, trade the markets, manage the risk and strip billions of dollars of cost out of the financial system.
Wonderful, except there is another side to this story that is becoming impossible to ignore because the people responsible for keeping the financial system alive are nervous.
Andrew Bailey, Governor of the Bank of England and Chair of the Financial Stability Board, has just warned the G20 that AI is becoming a threat to global financial stability, particularly as increasingly autonomous models develop sophisticated problem-solving and cyber capabilities.
A critical point in his argument is that AI is linking global systems from New York to London to Singapore, but each of these systems are run nationally which AI ignores.
This is where the AI debate gets much more interesting because the danger is not simply that artificial intelligence makes banks more vulnerable. The danger is that AI makes the entire financial system faster, more connected, more concentrated and more correlated at exactly the same time.
That is an explosive combination.
For example, the IMF warns that advanced AI dramatically reduces the time and cost required to discover and exploit software vulnerabilities.
Financial institutions depend upon shared cloud platforms, software, networks and payment infrastructure, which means that one weakness no longer threatens one bank. It threatens every institution using the same technology, and extreme cyber incidents could create funding pressures, solvency concerns, payment disruption, fire sales and a wider collapse of confidence.
The hacker used to attack the bank. AI attacks the banking system.
That distinction matters enormously because banking runs on confidence.
A cyberattack that steals a few million dollars is a loss. An AI-driven attack that simultaneously compromises several major institutions, disrupts payments and convinces customers that their money may no longer be accessible is a systemic event.
The Bank for International Settlements (BIS) makes the same point from another direction.
AI-powered attacks are becoming more sophisticated, scalable and difficult to detect, while dependence upon a small number of dominant technology providers creates concentration risk. Banking spent decades worrying about banks becoming too big to fail and is now quietly creating technology providers that are too interconnected to fail.
If the cloud computing infrastructure breaks down tomorrow, what will banks do?
This creates a double whammy of AI breaking banks systems whilst banks systems break down due to cloud services.
Then there is an even more interesting problem: what happens when every bank uses AI to make the same decisions?
Banks are embedding AI into credit underwriting, insurance, liquidity management, investment and trading. The Bank of England warns that common models, common datasets and common software components could cause institutions to misprice risk simultaneously.
That should sound familiar because the global financial crisis was not created because one bank misunderstood mortgage risk. It became systemic because thousands of institutions misunderstood the same risk at the same time.
AI could industrialise that behaviour.
Imagine thousands of institutions running similar models trained on similar information, identifying similar risks and responding within milliseconds. Something happens, the models detect danger, credit tightens, positions are sold, liquidity disappears and markets fall. The falling markets tell the models that risk has increased, triggering more selling, which creates more risk, which triggers more selling.
Humans call that panic.
Machines call it optimisation.
Research into AI, cloud computing and systemic risk has warned about exactly this problem for years, highlighting procyclicality, unknown unknowns and the possibility that intelligent systems optimise against the financial system itself.
The Bank of England now explicitly worries that AI-driven trading will create correlated positions, herding and amplified fire sales, while autonomous models could even discover that market stress creates profitable opportunities and behave in ways that increase instability.
That last point deserves far more attention because the financial industry has spent decades designing algorithms to maximise returns within defined constraints. Agentic AI changes the equation because the machine is increasingly deciding how to achieve the objective rather than simply executing predetermined instructions.
Tell an AI trading agent to maximise returns and somewhere inside that instruction lies a rather uncomfortable question: what happens if creating volatility is profitable? There is another vulnerability underneath all of this too … payments.
AI software is developing at extraordinary speed while the financial infrastructure underneath it was designed for a slower world. Payments, settlement, reconciliation, compliance and financial reporting remain tightly interconnected processes spanning multiple institutions and systems. Software can now change overnight while financial infrastructure still takes months or years to change.
This becomes critical as we move towards agentic commerce because AI will not merely analyse transactions. AI agents will initiate them. Machines will negotiate, buy, sell, borrow, invest and pay other machines, creating an economy operating at machine speed on financial infrastructure designed around human speed.
That is the contradiction sitting at the heart of banking’s AI revolution.
Banks desperately need AI because they cannot defend themselves against machine-speed attacks using human-speed defences, but banks cannot opt out. They have to put AI into cybersecurity because criminals have AI. They have to put AI into trading because competitors have AI. They have to put AI into lending because competitors will make better decisions faster. They have to put AI into payments because commerce is becoming autonomous.
The banking system is therefore entering an AI arms race in which everyone has to participate even though participation creates new systemic risks.
Meanwhile, another danger is building around the financing of AI itself.
The Bank of England notes that AI infrastructure is increasingly being financed through debt and that the scale of this financing is accelerating rapidly. If expectations around AI collapse, losses will therefore move beyond technology investors into credit markets and potentially into global financing conditions.
We therefore have several AI risks converging simultaneously: an AI investment bubble, growing debt financing of AI infrastructure, AI-driven cyberattacks, concentration amongst technology providers, correlated AI trading strategies, automated credit decisions, autonomous financial agents and legacy payment infrastructure struggling to support a machine-speed economy.
That is why the conversation has moved so quickly from what can AI do for banking? to what could AI do to banking?
We are heading towards a financial system where machines increasingly create the information, interpret the information, make the decisions and execute the transactions.
That does not mean AI will destroy banking. Quite the opposite. AI will become fundamental to banking because there is no credible alternative, but it does mean something much more important as the next financial crisis will not necessarily begin with a bad mortgage, a failing bank or a rogue trader. It could begin with an algorithm making a perfectly rational decision, followed milliseconds later by ten thousand other algorithms making exactly the same perfectly rational decision and, by the time the humans understand what happened, the machines will already have moved the money.
For more on this topic, click on the links below:
[3]: https://www.bis.org/review/r260608e.htm
[4]: https://www.bankofengland.co.uk/financial-stability-in-focus/2025/april-2025
[5]: https://ideas.repec.org/a/eee/jbfina/v140y2022ics0378426621002466.html
[6]: https://thefintechtimes.com/ai-softwares-real-bottleneck-is-payments-infrastructure
[7]: https://www.bankofengland.co.uk/financial-stability-report/2026/july-2026
Chris M Skinner
Chris Skinner is best known as an independent commentator on the financial markets through his blog, TheFinanser.com, as author of the bestselling book Digital Bank, and Chair of the European networking forum the Financial Services Club. He has been voted one of the most influential people in banking by The Financial Brand (as well as one of the best blogs), a FinTech Titan (Next Bank), one of the Fintech Leaders you need to follow (City AM, Deluxe and Jax Finance), as well as one of the Top 40 most influential people in financial technology by the Wall Street Journal's Financial News. To learn more click here...