#### Big Business This Week # Big Business This Week: Just How Big A Threat Is AI to Global Finance? By [Peter S. Green](/content/author/peter-green/index.html) Apr 17, 2026, 06:30 AM When Treasury Secretary Scott Bessent (above) and Fed chair Jay Powell jointly summon America’s top bankers to a meeting in Washington, you know it’s big. This time, it was over the specter of a new AI model created by Anthropic, called Mythos. If that sounds like a black-masked, sword-wielding supervillain, then it has a great name, because that’s what it sorta is. Before its public release, Mythos was let loose on the software running some of the biggest U.S. banks and it found flaws, faults and backdoors that had sat undiscovered, and uncorrected, for decades. Details of the closed meeting last week are scarce, but President Trump’s chief economic adviser, Kevin Hassett, told Fox News the Administration is **“taking every step we can to make sure that everybody is safe from these potential risks, including Anthropic agreeing to hold back the public release of the model until our officials have figured everything out. There’s definitely a sense of urgency.”** Anthropic is milking the supervillain for all its worth. It changed the model’s name from its original [Capybara](https://youtu.be/ME0uVtdN4vQ), an oversized, lovably cute South American hamster, and said it won’t release Mythos until a group of about 40 tech firms, including Apple, Amazon and Microsoft, get to use the model to scout for potential security breaches, and **“give good actors a headstart.”** It’s calling the group “Glasswing,” and offering $100 million in Claude usage credits to the project. But how serious a problem is it? Big Business This Week columnist Peter Green spoke with some cybersecurity experts to learn more. **“AI is becoming a very credible opportunity/threat to the financial system,”** said Jim Taylor, president and chief product and strategy officer at RSA Security, a respected internet security firm. AI is **“industrializing”** the speed and frequency of fraudulent transactions, way faster than banks and businesses can keep up. A recent survey by tech consulting firm SAS found that 47% of banks are caught in what the firm calls a **“trust dilemma,”** either deploying AI they haven’t fully validated or holding back AI they’ve validated and trust. More than the technology itself, the lack of people and procedures to manage the overnight growth of AI is what many regulators increasingly view as a systemic risk, the preparedness gap the Treasury and Fed are concerned about. The other big problem, said Taylor ,is the way AI is getting incorporated into the global financial system. **“It’s automation, it’s cheap, it’s better, it’s easier and more effective than a human. But if you start to rely on those AI models, the models themselves become a risk: what if those models are manipulated or they’re simply wrong?”** The big question, and what was probably behind the Washington convocation, said Taylor, was concern about how much knowledge banks have to understand and limit the risks of the AI they are incorporating at record speed. **“The Mythos announcement, there’s a fair amount of hype around it,”** said Jeff Williams, chief information security officer for Sigma 360, which provides compliance software to large financial institutions. Other AI models, some far simpler and less expensive to run than Mythos, have already found and exploited major security flaws, he said, putting hacking tools within reach of almost anyone with ill intent. **“That means that the barrier to an attack is suddenly lower,”** Williams said. **“How much lower is open for debate,”** but he noted, **“the fundamentals of security haven’t changed.”** That’s where security experts agree: credentials, identification and privilege go by many names, but the key to IT security is to ensure that only authorized users are entering your system. AI makes that a lot harder. It can gather hundreds of data points from confirmed users, everything from their syntax to their location or even their typing patterns, and fake their way into the system, erasing databases or diverting funds into fake bank accounts with fake invoices. **“We need to operate from a principle of least privilege,”** said Williams. **“We need to control our systems in a meaningful way so that they’re only doing the things that they’re supposed to do, that we can see that they’re doing them correctly.”** The lesson for defenders is that if attackers can move that fast, defenders have to move even faster, and that’s a problem for many large corporations, added Walker. **“Legacy compliance tools and security tools aren’t very well positioned for this increased pace. It takes a move to continuous compliance operations that leverage AI. That gap is not just technology, but also the people and process behind it, have you built a cadence that will allow rapid response to emerging threats.”** One interim solution, that may prove most durable, is to wall off key data on a server that is only reachable via what’s called an “air gap,” physically disconnecting from the internet, and using non-digital means to keep the data isolated. **“AI can’t circumvent physics,”** said Tony Hasek, CEO of [Goldilock Secure](http://goldilock.com/), a new NATO-standard air gap database protection device used by some U.K. utilities. **“The databases are physically disconnected,”** explained Hasek. **“Since the AI can only operate with Internet Protocol, if access to the databases is triggered by methods that don’t use the internet, there’s no way for the AI to break in.”** The weakest point in the armor around the financial system and its data are humans, says Shai Gabay, founder of Trustmi, a New York-based cybersecurity firm focused on countering AI identity fraud. AI gets better by the day at social engineering, detecting and duplicating the bits and pieces that make up a person’s online identity. **“Attackers understand how companies actually operate, and they’re using that to slip through gaps that tools were never designed to catch,”** Gabay said. **“The one thing that you cannot fix, it’s not software vulnerabilities, it’s the human beings, and that’s why social engineering is very effective because it’s targeting the human being behind the computer.”** The problems need to be solved soon, to prevent the erosion of public trust in the banking system. **“AI failures can undermine trust far faster than traditional risks, because flawed models, biased decision logic and ungoverned automation don’t remain isolated. By their very nature, these failures can cascade across customers, counterparties and markets in real time,”** said Stuart Bradley, senior vice president for fraud and security intelligence practices at SAS, the data security consulting firm. **“The potential for a sudden loss of confidence cannot be understated in today’s digital‑first financial system, where bad news spreads like wildfire and customers can move their money in an instant.”** That, says Goldilock’s Hasek, means globally regulators and financial institutions will need to work together to keep an eye on how AI is used. **“Everybody has to be willing to regulate it, because if one venue doesn’t regulate it,”** he said, **“we may be in deep trouble.”** _—Peter S. Green_ ## Big Businesses mentioned this week: - [$DIS ( ▲ 0.84% )](https://stocktwits.com/symbol/DIS?utm_source=beehiiv&utm_medium=email&utm_campaign=bbtw&_bhlid=52a43e276c9d6c5d5cb39a98cc697a5b708fcda6) - [$AMZN ( ▲ 0.48% )](https://stocktwits.com/symbol/AMZN?utm_source=beehiiv&utm_medium=email&utm_campaign=bbtw&_bhlid=568320a3047415efdcd37e8534829f9ab3a587f1) - [$GOOG ( ▼ 0.51% )](https://stocktwits.com/symbol/GOOG?utm_source=beehiiv&utm_medium=email&utm_campaign=bbtw&_bhlid=2f0a9a1d6f433bb9c1f3cbef95e1429a6e6608eb) - [$FLYYQ ( ▼ 10.36% )](https://stocktwits.com/symbol/FLYYQ?utm_source=beehiiv&utm_medium=email&utm_campaign=bbtw&_bhlid=e93d35e77011e0dee44a9a82224461398573322e) - [$TSLA ( ▼ 0.78% )](https://stocktwits.com/symbol/TSLA?utm_source=beehiiv&utm_medium=email&utm_campaign=bbtw&_bhlid=56a8b2efe0ec01c97ac145bd2ce8b74dfe96628b) - [$BIRD ( ▼ 35.79% )](https://stocktwits.com/symbol/BIRD?utm_source=beehiiv&utm_medium=email&utm_campaign=bbtw&_bhlid=9f69188b3e21e5a6e51a5c4bbb2aa39a5bd1ceca) - [$SNAP ( ▼ 0.33% )](https://stocktwits.com/symbol/SNAP?utm_source=beehiiv&utm_medium=email&utm_campaign=bbtw&_bhlid=6ea10bb966bc862b59aae84b24537945e8a1d927) - [$PTON ( ▲ 3.5% )](https://stocktwits.com/symbol/PTON?utm_source=beehiiv&utm_medium=email&utm_campaign=bbtw&_bhlid=e549c488072bb383cc80aa9e1b6c969c64a8eb2e)