Global banks are making more money than before. The International Monetary Fund doesn’t think that will save them from the next big threat.
The IMF says more of the world’s banks may fail in a doomsday scenario because of AI-related risks, even as their profits increase. Balance sheets look healthier while the worst-case outcomes look meaner.
What the IMF is actually worried about
The concern is not that a chatbot approves a bad mortgage. It is that advanced AI models make breaking into financial systems cheaper, quicker and more scalable.
In a May 7, 2026 blog post, the IMF argued that advanced models can cut the time and cost of identifying vulnerabilities. That, the fund said, pushes cyber risk up to systemic levels.
The post singled out Anthropic’s Claude Mythos as an example of a model that can speed up both the discovery and the exploitation of security gaps.
The bigger fear is correlation. A single bank getting hacked is a bad day for that bank. Many institutions failing at once, because they share the same weak points, is a bad day for payments, markets and public confidence in the whole system.
An IMF Note released in June 2026 spelled out where that correlation comes from. It flagged structural vulnerabilities in shared digital infrastructure, along with the advantage attackers gain from deploying AI quickly. The Note urged robust cross-border cooperation and steps to limit how far risk can spread once something breaks.
The same month, IMF Managing Director Kristalina Georgieva cautioned that frontier AI, if misused, could:
“destroy the financial system”
Georgieva also described the risk of an AI boom turning into a bust as low probability but high impact.
From firewalls to fallback plans
The IMF’s 2026 commentary includes a notable concession: the fund said cyber defenses in the financial sector “will inevitably be breached.” Instead of promising to keep every intruder out, the IMF is pushing institutions toward resilience planning: assuming the walls will fall and preparing to keep critical services running anyway.
In July 2026, the IMF widened the lens from hacking to trading. It warned that AI used in markets could synchronize risks across them, and it called on central banks to strengthen oversight and contingency measures.
The IMF has also highlighted concentration risk. A growing share of financial services runs through a small number of AI and cloud providers, which means a problem at one vendor can ripple through many institutions simultaneously.
Banks are leaning on AI for defensive jobs like fraud detection, yet the IMF’s view is that offensive AI capabilities are currently outpacing those defenses.
The fund has also noted that the fallout may not stay inside finance. Its warnings point to interconnected sectors like energy and telecommunications, which share infrastructure and dependencies with the banking system.
What this means for banks, regulators and investors
The most important takeaway for bank watchers is that profitability and resilience are not the same thing. By tying AI to the possibility of more bank failures in a severe scenario, the fund is effectively telling supervisors that their stress tests need to account for technology risk, not just interest rates and loan defaults.
For regulators, the direction of travel looks like more scrutiny. Central banks are being urged to beef up oversight, stress testing and contingency planning, and the IMF wants that work coordinated across borders rather than handled country by country.
The AI and cloud providers themselves are also in the spotlight. As the IMF keeps highlighting concentration, the companies at the center of financial infrastructure may face more pressure from regulators to prove their systems cannot become a single point of failure.
Georgieva herself called the AI boom-to-bust risk low probability. But the IMF is clearly signaling that, for the financial system, the cost of being wrong about AI could be very high, and the time to prepare is before the breach rather than after it.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.