Latest AI models could threaten the global banking system, financial officials warn

Imagine that new artificial intelligence models, such as those from Anthropic, are not just tools for answering your questions or generating text, but could actually threaten the stability of the global banking system. It sounds like something out of a movie, but these are the words of some of the world's top financial officials: the most advanced AI systems could expose vulnerabilities in banks' cybersecurity defenses, opening the door to unprecedented risks. We usually think of artificial intelligence as a neutral, useful technology – perhaps a bit unsettling for our jobs – but we rarely associate it with the security of our money or the trust we place in banks. Here’s the twist: the real threat is not that AI will take jobs away from tellers, but that it could bypass the digital defenses of the banks themselves. Claude Mythos, Anthropic's new model, has become the symbol of this concern. Financial regulators fear that these systems will be able to detect flaws in banking software much more quickly than security teams can. An executive at a major European bank recounted how, in an internal simulation, an AI model uncovered a vulnerability their team had never seen before, calling into question the bank's entire cybersecurity strategy. This incident set off alarm bells: If an AI model can find the key to breaking into digital vaults, what happens if this capability falls into the wrong hands? Banks have always invested millions in firewalls and defense systems, but now they are facing an adversary that learns, adapts, and never tires. A sobering fact: according to a recent report, 70% of the world's banks have already included AI attack scenarios in their stress tests. Until just a few months ago, these risks were not even considered. The way authorities view the problem is also changing: the threat no longer comes solely from lone hackers or organized groups, but from intelligences capable of independently discovering the weak points of an entire system. Some argue that the real race is not between banks and hackers, but between those who develop AI for defense and those who use it for attack. What if, tomorrow, AIs started collaborating with each other, beyond human control? Today, this question is no longer science fiction. But there is a dissenting voice: some cybersecurity experts point out that every new technology brings with it an initial wave of panic. When computers were introduced into banks, there were fears that branch offices would collapse. Today, they say, the real challenge is to successfully integrate AI into defense systems before criminals do so. In short, the biggest threat is not AI itself, but the gap between its learning speed and our ability to adapt. If you thought artificial intelligence was just about automation, now you know that it could become the key to disrupting the entire banking world. If this perspective has changed your view of what is really at risk with AI, on Lara Notes you can indicate it with I'm In: it's not a like; it's a way of saying that this concern now concerns you personally. And if you find yourself telling someone how AI could uncover banks’ vulnerabilities before humans do, on Lara Notes you can tag that conversation with Shared Offline: it’s a way of saying that certain topics are worth remembering, even beyond the screen. This Note comes from the Financial Times and has saved you at least eight minutes of reading.