Bessent Summons Bank CEOs Over Anthropic AI Cyber Risks
Scott Bessent Anthropic Mythos cyber risks 2026
United States Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell convened an urgent meeting with the nation’s most powerful bank CEOs on Tuesday, April 7, 2026. The high-level gathering at the Treasury Department focused on the unprecedented cybersecurity risks posed by Anthropic’s latest artificial intelligence model, known as Claude Mythos. Treasury officials warned the leaders of Bank of America, Citigroup, Goldman Sachs, Morgan Stanley, and Wells Fargo that the new system possesses a unique ability to automate the discovery and exploitation of software vulnerabilities.
According to reporting from Bloomberg News, the meeting was arranged on short notice after intelligence suggested that the model could destabilize the global financial infrastructure. The government remains deeply concerned that such advanced AI capabilities could eventually fall into the hands of hostile state actors or sophisticated criminal syndicates.
The model at the center of the controversy, Mythos, has already demonstrated a terrifying level of proficiency in identifying flaws that have remained hidden for decades. Anthropic recently revealed that the AI successfully uncovered a critical vulnerability in the OpenBSD operating system that had gone undetected for twenty-seven years. Furthermore, the model identified a significant flaw in the FFmpeg video tool that had escaped five million automated security tests.
This capability has led to a massive shift in how the financial sector views its defensive posture against automated attacks. As detailed by The Economic Times, Secretary Bessent urged the attending CEOs to expedite the patching of their internal systems before similar AI tools become widely available to malicious hackers. The absence of JPMorgan Chase CEO Jamie Dimon, who was unable to attend the summit, did not diminish the gravity of the discussions regarding the systemic importance of these technological threats.
In response to the growing alarm, Anthropic has implemented a highly restricted release strategy for the Mythos model and launched a defensive initiative called Project Glasswing. This partnership involves major technology firms such as Amazon, Apple, Google, and Microsoft, along with several systemically important financial institutions. The goal of Project Glasswing is to use the model’s offensive capabilities to find and fix security holes before they can be exploited by outsiders.
Anthropic has committed approximately $100 million in usage credits to support these defensive efforts and has donated $4 million to various open-source security organizations. According to updates from the Times of India, the company has no current plans for a broad public release of the model due to the inherent dangers. This cautious approach mirrors the prudence of federal regulators who now view AI-driven cyberattacks as one of the greatest risks to the stability of the American economy in 2026.
The financial fallout from the emergence of Mythos has already been felt across global markets, including a $2 trillion stock wipeout in the technology and cybersecurity sectors earlier this year. Investors reacted with panic following a configuration error in late March that allowed some details of the model’s capabilities to leak online. Treasury Secretary Bessent and Chairman Powell emphasized that the stability of the banking system depends on a proactive and coordinated response between the private sector and the federal government.
While the administration continues to support the development of domestic artificial intelligence, it is clear that the focus has shifted toward rigorous oversight and the prevention of catastrophic digital failures. As reported by various financial news outlets, the Tuesday meeting signals the beginning of a new era of regulatory intervention in the intersection of finance and advanced machine learning. The Treasury Department is expected to issue further guidance to all systemically important banks regarding the integration of defensive AI protocols in the coming weeks.
