A wave of unease swept through Washington this week as financial leaders gathered behind closed doors, confronting what some fear could be a digital earthquake: the rising Anthropic AI cyber risk.

Treasury Calls High-Stakes Meeting

Scott Bessent convened top executives from America’s largest banks for an urgent meeting, according to reports, as concerns mounted over the cybersecurity implications of a powerful new AI system.

Among those present was Jerome Powell, who joined discussions at Treasury headquarters following the emergence of Anthropic’s new AI model, Claude Mythos—technology the company itself warns could expose unprecedented vulnerabilities.

AI Model Raises Alarm Bells

Anthropic recently disclosed that its upcoming Claude Mythos model has capabilities that surpass nearly all humans in identifying and exploiting software weaknesses.

The warning came after a leak of the model’s code prompted the company to publish a stark blog post, cautioning that the consequences for economies, public safety, and national security could be severe. The message landed like a thunderclap across both Silicon Valley and Wall Street.

Banking Giants in the Room

The meeting brought together leaders of systemically important banks—institutions whose failure could ripple across the global economy.

Attendees reportedly included David Solomon of Goldman Sachs, Brian Moynihan of Bank of America, Jane Fraser of Citigroup, Ted Pick of Morgan Stanley, and Charlie Scharf of Wells Fargo.

Jamie Dimon of JPMorgan Chase was invited but did not attend.

In his annual letter to shareholders released the same week, Dimon underscored the urgency, warning that cybersecurity remains one of the banking sector’s greatest threats—and that artificial intelligence is likely to amplify those dangers.

Restricted Release Signals Unusual Concern

Anthropic has taken the rare step of limiting access to Claude Mythos, marking the first time the company has restricted a product rollout.

Early access has been granted only to a small circle of major technology players, including Amazon, Apple, and Microsoft, along with networking firms Cisco and Broadcom. The Linux Foundation has also been included.

Hidden Flaws, Long Buried

According to Anthropic, the AI system has already uncovered thousands of previously undetected vulnerabilities in widely used software—some dating back as far as 27 years. Many of these flaws had gone unnoticed by developers and security monitors alike, raising fears that malicious actors could exploit similar tools to crack passwords or bypass encryption safeguards.

Government Scrutiny Intensifies

The meeting follows the U.S. government’s recent designation of Anthropic as a supply chain risk—a classification the company is now contesting in court.

Despite the high stakes, key players—including the Federal Reserve, Anthropic, and the major banks—declined to comment publicly. The Treasury Department also did not respond to requests for clarification.

A Digital Fault Line Emerges

The Anthropic AI cyber risk debate is rapidly evolving into a defining issue of the AI era—where innovation moves at lightning speed, but the guardrails struggle to keep pace. Like a master key that can unlock both doors and vaults, the same technology promising progress may also expose the world’s most sensitive systems to unseen threats.