On Tuesday morning, Jamie Dimon’s voice had a strange quality that you had to pay close attention to. The CEO of JPMorgan Chase, who is typically a picture of grizzled certainty, sounded almost cautious when he spoke during the bank’s quarterly earnings call. The topic was artificial intelligence, specifically Anthropic’s new Mythos model. He told analysts, “AI’s made it worse, it’s made it harder,” which is a straightforward statement that sticks because it doesn’t attempt to soften anything. He continued, “It does create additional vulnerabilities and perhaps better ways to strengthen yourself in the future.” Perhaps in the future, that qualifier will reveal the true story. Hedging is not Dimon’s specialty.
Since Anthropic made the preview available to a limited number of companies on April 7, JPMorgan and a few other major institutions have been testing Mythos. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell called an emergency meeting of the major bank executives on the same day it fell. You can tell how seriously Washington is taking this by looking at that particular detail. The most influential figures on Wall Street are not called in on a Tuesday afternoon to discuss productivity tools.
When a reporter questioned Dimon about the model, he appeared to be pointing directly at Anthropic’s warning that Mythos had already found thousands of vulnerabilities in corporate software. “I think you read exactly what is it,” he replied. It indicates that many more vulnerabilities need to be addressed. Go over that twice. He’s not in a panic. With the dry weariness of someone who has watched cybercriminals become more sophisticated over the past 20 years, he is observing that the floor has just shifted once more.
Mythos is unique in that it was intended to be useful. Identify software bugs, fix them, and strengthen the system. Anthropic’s CEO, Dario Amodei, has characterized the situation as dangerous because the same ability that shields can also reveal. To its credit, Anthropic decided to release Mythos on a limited basis rather than publicly. Nevertheless, there is a clear fear in the financial community: a tool that identifies industrial-scale vulnerabilities is, by definition, also a tool that maps the terrain for those with malicious intent.
Later, during a livestreamed Anthropic event, Dimon urged Amodei to provide more details, saying, “The right thing for Dario and Anthropic is to lay it out, give people a chance to study it, understand the vulnerabilities, come up with plans so we can handle it.” One CEO is effectively asking another to slow down and open the hood, which is a startling request. Dimon seems to be concerned about more than just JPMorgan. The exchanges, clearinghouses, small banks, routers, and out-of-date passcodes that nobody updates until something breaks are the connective tissue that worries him.
The chief financial officer of JPMorgan, Jeremy Barnum, was more measured but equally direct. According to him, these tools are capable of identifying vulnerabilities and can be used in attack mode. Recent developments have exacerbated a trend that the industry was already aware of.
David Solomon acknowledged on Monday that Goldman Sachs is also testing Mythos, but he did not provide further details. Speaking at a roundtable of the Financial Stability Oversight Council, Federal Reserve Vice Chair Michelle Bowman described the event as a reminder of how quickly artificial intelligence is developing and how ill-suited current risk frameworks are to it. She proposed that regulators would have to make last-minute changes to the rulebook.
This, in Dimon’s opinion, is a passing phase in a longer arc. Perhaps he is correct. As this develops, however, it’s difficult to avoid the impression that, despite the threat outgrowing them, the banks have entered a time when the old hygiene—strong passwords, patched routers, and watchful teams—matter more than ever. On both sides of the wall, the tools are becoming more intelligent. On Tuesday, nobody in the room seemed prepared to respond to the question of which side adapts more quickly.
