The most powerful banker in America just said the AI model built to make systems safer made them 10 times more vulnerable.
The Summary
- JPMorgan CEO Jamie Dimon told Bloomberg that Anthropic's Mythos AI model increased cybersecurity risks "10-fold" by exposing vulnerabilities banks didn't know existed
- Dimon has long cited cyber as JPMorgan's biggest risk, making his specific call-out of Mythos particularly notable
- The warning follows recent safety testing incidents at both Anthropic and OpenAI that raised questions about frontier model control
- This marks a shift from abstract AI risk discussions to a major bank CEO naming a specific model as a clear and present threat
The Signal
Jamie Dimon doesn't do hyperbole about technology. The man who runs a $3.9 trillion balance sheet just went on record saying a single AI model made cyber risks 10 times worse. Not "could increase risks" or "raises concerns." Made them worse. Past tense. Already happened.
The specificity matters. Dimon told Bloomberg TV that "AI created vulnerabilities that we didn't know about" at the JPMorgan Tech Stars Conference in London. This isn't about theoretical attack vectors or white papers. This is the CEO of America's largest bank saying his security teams are now defending against threats that didn't exist before Mythos shipped.
"We always worried about cyber before these things. The risks went up 10-fold after Mythos."
The timeline is crucial here. Anthropic and OpenAI both disclosed separate safety testing incidents in recent months that raised control questions. Dimon's comments suggest those weren't isolated lab incidents. They were canaries. The vulnerabilities these frontier models expose aren't edge cases. They're systemic.
What makes this different from the usual AI safety discourse: Dimon is talking about realized risk, not hypothetical doom scenarios. Banks test their systems constantly. They know what normal threat escalation looks like. A 10x jump is not normal. That's the kind of number you get when the rules of the game change.
Key implications:
- Advanced AI models are offensive cybersecurity tools first, defensive tools second
- The vulnerability discovery capability outpaces the patching capability by an order of magnitude
- Financial institutions are facing threats they can't fully map yet
The broader context: AI researchers, tech executives, and lawmakers have renewed calls for stronger international cooperation to reduce catastrophic AI misuse. Dimon's statement gives those calls teeth. When the CEO who guards the financial system's plumbing says a model created a 10-fold risk increase, regulators listen differently.
The Implication
If Mythos exposed vulnerabilities JPMorgan didn't know it had, every other financial institution has the same blindspots. The question now is whether security teams can discover and patch those holes faster than attackers can weaponize them. Early indicators say no.
Watch for increased regulatory pressure on frontier AI labs to disclose model capabilities before release, not after. Dimon's comments give banking regulators a clear mandate: if AI models can 10x cyber risk overnight, the approval process can't stay in 2023. The window between "this model can find new exploits" and "bad actors are using this model to find new exploits" is measured in weeks, not quarters.