Anthropic exposes vulnerabilities in G20 banks as Meta pours up to $145 billion into AI

Claude Mythos works as a tool for finding vulnerabilities in critical infrastructure. Anthropic unveiled the model in April, positioning it as a solution for cybersecurity and risk assessment in financial systems. The results were impressive enough to force governments to take it seriously.

What Claude Mythos found

The model uncovered thousands of critical problems in major operating systems, browsers and web applications. When the company asked it to write working exploits for these vulnerabilities, Claude succeeded on the first attempt in 83% of cases.

"It looks like the company has found a way to break the entire world of cyber risk" —

Andrew Bailey, Bank of England

The result was serious enough to catch the attention of Andrew Bailey, Governor of the Bank of England and chair of the Financial Stability Board. Bailey formally asked Anthropic to brief the FSB, an organization that includes financial authorities from the US, UK, Canada, France, Germany, Japan, Saudi Arabia, Australia and China.

But access to Mythos is strictly limited: only 40 organizations have it, mostly American (Amazon, Microsoft, JPMorgan Chase). The White House asked that the group not be expanded. Russian banks will learn about critical vulnerabilities only from news reports.

Meta goes all in on AI

On May 20, Mark Zuckerberg announced the layoff of 8,000 employees (10% of Meta's staff). The company had just posted record quarterly results: revenue of $56.3 billion and profit of nearly $27 billion. This is not a crisis-driven cut but a deliberate restructuring around AI.

In an internal memo, Zuckerberg wrote that AI is "the most important technology of our time" and that the companies leading its development will "define the next generation." The conclusion: focus only on the most capable people and invest enormous sums in computing power. Alongside the layoffs, 7,000 people are being moved into AI development roles.

Meta's 2026 capital expenditure will reach $125-145 billion, more than Estonia's GDP.

• Electricity consumption of AI infrastructure
• Retraining 7,000 employees for AI roles
• Abandoning balanced growth in favor of concentration in a single area

CFO Susan Li admitted in an interview that "the company frankly does not know what the optimal size will be in the future."

What this means

The two stories merge into one pattern: the concentration of technological power is accelerating faster than regulators and traditional organizations can adapt. One American private company now controls a tool for finding breaches in banking infrastructure. Most banks run on 30-year-old COBOL systems. Regulators lag by a year and a half. Employees cannot retrain fast enough. This is either the most interesting decade in the history of technology or the most dangerous.

*Meta is recognized as an extremist organization and is banned in Russia.