Artificial intelligence is becoming part of the everyday operating system of financial services, from fraud detection and identity checks to customer service and transaction monitoring.
While AI is useful for banks, it also raises a question for risk officers and regulators. What happens when automated systems help move fear as quickly as money?
A bank run occurs when depositors lose confidence that a bank can meet withdrawals and rush to move their money elsewhere. If too many customers demand cash at once, a liquidity problem can metastasize into an institutional, and ultimately system-wide, crisis.
The next run may not begin with the familiar warning signs of bad loans, souring credit or a weak securities portfolio. It could spark with fraud, compromised credentials, a payments outage, or automated instructions that convince customers, companies or their software agents to move money before facts are clear.
Fraud Can Become a Confidence Event
The PYMNTS Intelligence report “2025 State of Fraud and Financial Crime in the United States” showed why fraud belongs in the bank-run conversation. Unauthorized-party fraud, driven by credential theft and account takeovers, represented 71% of total incidents and dollar losses last year, up from 48% of incidents in 2024.
That is not the same thing as saying fraud will cause a bank run. It means the mechanisms that could damage confidence are becoming more digital, more automated and harder to isolate.
Advertisement: Scroll to Continue
A realistic scenario would not require a cinematic cyberattack. It could start with a wave of account takeovers, false balance information, blocked payment access or fraudulent transfer attempts affecting a visible group of customers. If customers believe their money is unsafe or inaccessible, they may move deposits quickly. If corporate treasury systems or AI-enabled financial agents are programmed to respond to risk alerts, they may do the same without waiting for human judgment.
Automation changes the risk profile. A human depositor may pause, call the bank or wait for confirmation. A rules-based treasury tool, compromised account script or AI agent may execute instructions immediately. At scale, those actions could turn a fraud incident into a liquidity problem.
Financial institutions are already under pressure from faster payments and more varied payment types. PYMNTS Intelligence found that 46% cited increased payment speed as a fraud-management challenge, while 41% cited the expansion of payment types and currencies, including peer-to-peer, instant transactions and cryptocurrencies.
Those faster rails are valuable in normal conditions. Under stress, they leave less time to distinguish legitimate withdrawals from fraud-driven activity or automated overreaction.
The report also showed that fraud’s damage is not limited to losses. Half of financial institutions said fraud hurt customer loyalty, 44% cited damage to brand and reputation, and 47% reported operational disruptions or failures. These are the ingredients of a confidence problem that translate into deposit flight.
Identity Becomes a Stability Tool
Banks are spending more because the risk is no longer only transactional. PYMNTS Intelligence found that 68% of financial institutions increased fraud-detection spending year over year, while 46% said fraud schemes had become more sophisticated.
In an AI-driven environment, banks need to know whether a transaction is being initiated by the actual customer, a criminal with stolen credentials, a synthetic identity, a hijacked bot or an authorized software agent acting on flawed information.
Strong identity protocols can slow unauthorized withdrawals, detect unusual behavior and prevent fraudulent activity from being mistaken for legitimate deposit flight. They can also help banks communicate with customers more credibly because the institution can show that access, authentication and transaction controls are still functioning.
The International Monetary Fund warned this month that AI-driven cyber risk should be treated as a financial stability issue because attacks can affect payment systems, confidence and liquidity at the same time.
The read-across from Silicon Valley Bank has some merit. SVB’s 2023 collapse was not caused by AI or fraud. It was driven by interest-rate risk, liquidity pressure and depositor concentration. However, it showed that digital channels and rapid communication can compress the timeline of a run. AI could shorten that timeline further if automated systems respond to risk signals before people can verify them.
The practical lesson for banks and FinTechs is that capital and liquidity still matter. But operational resilience, cyber redundancy, identity verification and customer communication now sit closer to the center of financial stability.
The next bank run may still come from bad loans. The added consideration is that it starts with compromised access, automated transfers and customers who decide that the safest response is to leave first and ask questions later.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
