Chill! Jamie Dimon did not say that artificial intelligence is going to destroy the world. He only said something potentially more consequential for the financial system: the cybersecurity risks created by advanced AI have, in his assessment, increased tenfold since Anthropic’s Mythos model emerged. “AI created vulnerabilities that we didn’t know about,” the JPMorgan CEO said in a Bloomberg interview.
The number — tenfold — is Dimon’s own estimate; but dismissing it would miss the point, because Dimon runs one of the world’s largest financial institutions, where cyber risk is not an abstract technological concern. It is a direct threat to payments, customer information, trading systems, and the infrastructure on which modern finance depends. His warning arrives as advanced AI systems are increasingly capable of acting on their own, including during safety testing in ways their developers did not fully anticipate.
And the AI debate is changing.
There was a time when we asked whether machines could make mistakes while producing information. The question nowadays is whether increasingly autonomous systems can do things before humans understand exactly what they are doing. We have developed systems that managed to write malicious code, probe networks, exploit vulnerabilities, or take unauthorized actions. This is fundamentally different from a chatbot that gives a wrong answer.
And the unfortunate part is that the companies developing these systems are discovering some of these behaviors while the systems are already being built and deployed. Anthropic’s own safety testing of Mythos reportedly found models accessing the internet and taking unauthorized actions. The problem therefore isn't simply malicious humans using AI. It is also the possibility that a sufficiently capable system can pursue an assigned objective through methods its creators did not anticipate.
This is a critical point where markets may experience their own AI correction, and it isn't because artificial intelligence will cease to be profitable; in fact, the opposite is likely true. The correction could occur as investors start to consider questions that have not been consistently factored into AI valuations. These questions include: What liabilities are associated with autonomy? Who is responsible when an autonomous system causes a security breach? Who bears the cost when an AI agent conducts an unauthorized transaction? Who is held accountable when a model exploits a vulnerability that its developer was unaware of? Additionally, how much capital must companies set aside to cover risks that regulators have yet to clearly define?
For years, the AI race has rewarded speed. Companies compete to build larger models, acquire computing power, secure chips, attract talent, and reach consumers before their competitors do. Regulation has struggled to keep pace. There's a political problem here, but also an economic one that strikes deep in the soul of the market: the private rewards of moving quickly are immediate, while some of the potential costs of moving recklessly are delayed, distributed, and difficult to quantify.
So, Dimon's warning cannot be dismissed because it brings those hidden costs into the language of finance.
If AI risk becomes sufficiently visible, investors may begin demanding something beyond faster models and larger data centers. They may demand evidence of containment, cybersecurity, insurance, human oversight, and legal accountability. The companies that can demonstrate those protections may eventually command a premium. Those that cannot may discover that technological capability alone is no longer enough to justify their valuations.
That would not be an anti-AI correction. It would be something more rational: a market beginning to price the risks that enthusiasm ignored. And the irony is that this may be exactly what the AI industry needs. Fear is not a substitute for regulation, and panic is not a safety system. But neither is corporate optimism. If machines are becoming capable of acting independently, then society has to curb its enthusiasm about AI and decide how much independence we are willing to permit before we know how to control the consequences.
No comments:
Post a Comment