Willie Sutton, the notorious bank robber, famously quipped he targeted banks “because that’s where the money is.” But a new kind of heist is looming, one not involving masks and guns, but lines of code and artificial intelligence. The threat isn’t from lovable rogues anymore; it’s from the relentless advance of AI itself.
For years, the possibility of AI-driven cybercrime felt like a distant warning. Now, that future appears to be arriving with alarming speed. Banks worldwide are bracing for a new wave of attacks, powered by the very technology designed to protect us.
At the heart of this concern lies “Mythos,” Anthropic’s latest AI model – a system so powerful it’s currently unavailable to the general public. Its creators believe unleashing such capability without safeguards would be reckless, a digital back door blasted wide open to the world’s financial vaults.
Internal tests have revealed a chilling truth: Mythos has uncovered thousands of critical security flaws across virtually every operating system and web browser. Some of these vulnerabilities have remained hidden for decades, representing “zero-day” exploits – immediate, catastrophic risks demanding instant attention.
Anthropic isn’t keeping this knowledge locked away. They’ve partnered with tech giants like Microsoft, Amazon, Apple, and Cisco, providing access to Mythos and committing substantial resources – over $100 million – to identify and patch these dangerous weaknesses. A select group of banks, primarily in the US, are also involved in this critical defensive effort.
However, a troubling gap exists. As of now, banks in Australia, the United Kingdom, and Europe haven’t been granted access to Mythos. Adding to the urgency, reports surfaced this week of unauthorized users briefly gaining access to the model, though the intent behind this breach remains unclear.
Should everyday people be alarmed? Fortunately, robust financial protections are in place in many countries. For example, Australia’s Financial Claims Scheme insures deposits up to A$250,000, and regulators actively pursue fraudulent transactions. A frantic rush to empty bank accounts isn’t warranted, but vigilance is.
The immediate response should be simple: keep your devices updated. Regularly install the latest operating system and banking app updates, as more will undoubtedly be released in the coming weeks and months. And, as always, remain acutely aware of phishing attempts via email and SMS – the oldest trick in the cybercriminal’s playbook remains remarkably effective.
Mythos highlights a fundamental truth: defense is exponentially harder than attack. Software, by its very nature, is incredibly complex, making it virtually impossible to eliminate all vulnerabilities. This creates a perpetual arms race, a relentless pursuit of flaws before they can be exploited.
The recent rollout of the European Union’s age verification app serves as a stark example. Within hours of its launch, security experts identified exploitable vulnerabilities, demonstrating how easily even well-intentioned software can be compromised.
In critical applications, rigorous mathematical proof of software integrity is the ultimate goal. Initiatives like the Beneficial AI Foundation’s “moonshot” project, aiming to verify the security of the Signal messaging app, represent a promising, though currently rare, approach.
Perhaps, ironically, further advancements in AI will eventually provide the tools to overcome these challenges. But for now, the landscape is shifting, and the stakes – the security of the global financial system – have never been higher.
