• The Bank of England's Financial Stability Report, published on 7 July 2026, says recent frontier AI progress significantly increases the risks to financial stability from cyber and operational vulnerabilities.
• The Bank says frontier AI models are increasingly capable of finding and exploiting software vulnerabilities at greater scale and over multiple stages, though AI also offers chances to improve cyber defence.
• The IMF's April 2024 analysis found modest, somewhat persistent deposit outflows at smaller US banks after attacks, and no significant cyber runs so far.
• The two documents answer different questions: the Bank covers how attack risk is changing, the IMF covers how an attack could propagate through funding. Linking them is an inference.
• Compliance teams can map single-vendor dependencies in payments and customer access, and treat a cyber incident as a confidence event as well as a downtime event.

Cyber risk to banks is again a central-bank concern. The Bank of England's July 2026 Financial Stability Report ties it to frontier AI, and earlier IMF analysis describes how an attack could turn into a funding problem through lost depositor confidence. A trade-press item from Finextra, which prompted this piece, summarises both warnings. This analysis rests on the two primary documents instead.

What the Bank of England said in July 2026

The Bank of England's Financial Stability Report, published on 7 July 2026, says recent progress in frontier AI presents a significant increase in the risks to financial stability from cyber and operational vulnerabilities. It states that frontier AI models are increasingly capable of identifying and exploiting software vulnerabilities at greater scale and over multiple stages, and that operational risks are likely to increase as AI accelerates vulnerability discovery and exploitation. It also acknowledges that frontier AI offers opportunities to improve cyber defence.

What the IMF found on deposits

The IMF's April 2024 analysis, drawn from its Global Financial Stability Report, found modest but somewhat persistent deposit outflows at smaller US banks after cyberattacks, and no significant cyber runs so far. It describes a chain in which eroded confidence and interrupted critical services, such as payment networks, can spill over to other institutions and create funding problems that could threaten solvency. The analysis is about two and a half years old, so it shows the mechanism rather than current conditions.

Where the two documents connect

The two documents answer different questions, and the Bank of England text reviewed for this piece does not discuss deposits. The Bank describes how the likelihood and impact of attacks may change; the IMF describes how an attack that lands could propagate through funding. Read together, they suggest compliance teams should ask whether faster, AI-assisted exploitation shortens the time between a vulnerability and a customer-visible outage. That is an inference, not a finding of either institution.

What this means for compliance teams

Teams can treat a cyber incident as a confidence event as well as a downtime event. Concretely, they can map which payment and customer-access services depend on a single outside provider, test the Bank's question of whether current deep cyber recovery capabilities remain sufficient, and rehearse customer communications before an outage. The IMF's example of a 2023 ransomware attack on a cloud provider, which caused simultaneous outages at 60 US credit unions, shows why shared vendors matter.