The next financial crisis might not be caused by humans, but by AI?
A recent bombshell—the new AI model Mythos from AI giant Anthropic—has sent shockwaves through Wall Street.
01
On the very day the Mythos model was released, an extremely rare scene occurred:
U.S. Treasury Secretary Bessent and Federal Reserve Chair Powell set aside all routine agendas and urgently convened the CEOs of all major Wall Street banks for a closed-door meeting at the Treasury.
U.S. Secretary of the Treasury Scott Bessent (left) and Federal Reserve Chair Jerome Powell (right)
Attendees included Jane Fraser, CEO of Citigroup; David Solomon, CEO of Goldman Sachs; Ted Pick, CEO of Morgan Stanley; and other key leaders of the U.S. financial system.
Keep in mind that the last time the Federal Reserve and the Treasury deployed an emergency summons of this magnitude was likely during the 2008 subprime mortgage crisis.
This time, their goal is not to "raise funds to rescue the market," but to ensure banks recognize that the newly released AI model, Mythos, despite its recent launch, could pose systemic risks even greater than the subprime mortgage crisis of the past.
Powell's personal attendance sent a very strong signal to the outside world:
This is no longer a technical discussion, but a top-tier issue concerning financial stability and even national security.
But this was not just a unilateral alert from the United States—market anxiety quickly spread across the Atlantic. Almost in perfect synchrony, the Bank of England and the Financial Conduct Authority (FCA) urgently initiated consultations with the National Cyber Security Centre (NCSC).
The Bank of Canada also separately summoned domestic banks and financial institutions for the highest-level risk advisory.
The consecutive actions by the central banks of the U.S., U.K., and Canada within a 10-day period are unprecedented in the history of AI regulation.
02
How powerful is Mythos really?
Anthropic claims that Mythos has extraordinary capabilities to uncover hidden vulnerabilities in the software running global banks, power grids, and government institutions.
OpenAI CEO Sam Altman compared Mythos to "dropping bombs while selling $100 million bomb shelters."
In simple terms, it can identify and exploit vulnerabilities in all major operating systems and all major web browsers, causing devastating attacks without human intervention.
During several weeks of internal testing, Mythos autonomously discovered "thousands" of zero-day vulnerabilities. It also uncovered a 27-year-old hidden vulnerability within the highly secure OpenBSD system.
At the same time, it discovered a 16-year-old vulnerability in another video-processing tool—a flaw that even the most advanced automated testing tools, despite running millions of times, had completely missed.
Previously, the top human hacker teams could find at most 100 such critical vulnerabilities per year. Mythos produces 10 to 100 times more output than the top human teams and reduces the time to develop exploit code from “weeks” to “hours.”
Typically, the "firewall" of financial institutions is among the most heavily fortified areas globally, protected by top-tier security experts in a moat-like arrangement.
But Mythos demonstrates an asymmetric destructive force that renders financial "firewalls" powerless.
This raises two critical questions:
First, could Mythos be used by malicious actors to attack the financial system? Will the money in our bank accounts still be safe in the future?
Second, could similar models be weaponized by certain countries in the future, for example, to target institutions and businesses of adversarial nations?
Think about it—it’s terrifying.
Even the developer Anthropic is now a bit wary and has voluntarily announced that Mythos will not be open to the public, but will only be provided to a select few institutions for testing.
03
The pace of change in the times is so rapid that it leaves people unable to react.
Researcher Tan Shaoyou from ZhiGu Trend believes that the evolution of AI is giving rise to a divergence beyond imagination—not only a gap in cognition, but also a massive disparity in power.
While ordinary people still view Doubao and DeepSeek as chatbots, the large models mastered by top global AI teams are already capable of undermining the foundations of the financial system.
A recent Morgan Stanley report highlighted that AI differs fundamentally from previous technological revolutions in one key aspect: speed.
It took humans sixty years to complete the Industrial Revolution, thirty years for the internet, and just two months for ChatGPT to reach one hundred million users. The curve of technological diffusion is steep enough to take your breath away.
The problem is that the pace of technological advancement has created a historical chasm between it and the rates at which regulatory legislation, societal defenses, corporate security measures, and personnel training are evolving.
As National Economic Council Director Kevin Hassett frankly admitted in the interview:
Even the White House and government levels have not yet fully prepared for this explosive pace of iteration.
Goldman Sachs CEO Solomon described this threat as “highly alert,” as the current defense threshold has shifted from “continuously strengthening firewalls” to “must counter the strongest AI with the strongest AI.”
This means the gap between the pace of technological advancement and societal adaptation has been stretched to historical extremes. Friction will be more intense than ever before.
As Musk said: "In the next 3 to 7 years, it may be very difficult for the average person."
In response, ordinary people must embrace the mindset of lifelong learning. The ability to continuously define and articulate complex problems is the greatest moat in the AI era.
At the same time, AI can replace your skills, but it cannot replace your assets. Building income structures that don’t rely on “trading time” has become even more important.
This article is from the WeChat public account "ZhiGu Trend," authored by Bu Yu.