Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell have convened emergency meetings with major Wall Street executives to address cybersecurity risks posed by Anthropic's latest artificial intelligence model, known as Mythos. The high-level discussions follow reports that the unreleased system successfully identified software vulnerabilities dating back 27 years, flaws that had previously evaded detection by human experts and legacy security systems.
According to multiple reports, the meetings were prompted by concerns that Mythos possesses capabilities far exceeding current industry standards. While Anthropic has restricted public access to the model, citing safety risks regarding potential misuse by malicious actors, similar capabilities may already be within reach of other developers. The model's ability to penetrate existing cyber defenses has triggered a significant reaction in financial markets, with cybersecurity stocks experiencing notable volatility.
In response to the threat landscape, Anthropic has initiated a limited rollout of Mythos to select technology partners as part of a cybersecurity defense initiative dubbed Project Glasswing. Companies including Amazon, Apple, Microsoft, CrowdStrike, and Palo Alto Networks will utilize the model to test and strengthen their own security infrastructure. This move aims to leverage the AI's offensive capabilities for defensive purposes before broader deployment.
The urgency of the situation was underscored by separate meetings held between Bessent, Powell, and heads of major U.S. banks to discuss the potential impact on financial stability. Similar concerns were echoed internationally, with the Bank of Canada and Canadian financial firms holding parallel discussions regarding the risks associated with the new AI technology. Financial Times reported that the Treasury Secretary specifically called in U.S. bank CEOs to address the cyber risks, highlighting the cross-border nature of the threat.
Market reactions have been swift. Investors appear increasingly spooked by the prospect of an AI-driven "day of reckoning" for software security, leading to a selloff in cybersecurity stocks. Palo Alto Networks' shares led the decline, reflecting broader anxieties that existing security architectures may be insufficient against such advanced AI agents. Analysts note that the revenue projections and updates from Anthropic have revived fears regarding the rapid pace of the cyber arms race.
Amidst these security concerns, Anthropic continues to expand its operational capacity. The company recently secured expanded chip agreements with Broadcom and Google, addressing chronic shortages in AI computing hardware. These deals, which include the production of future versions of Google's AI chips and expanded supply from Broadcom, come as Anthropic reports annualized revenues hitting $30 billion. Additionally, sources indicate the company is exploring the possibility of designing its own custom AI chips to further reduce reliance on external suppliers.
While Anthropic describes Mythos as "by far the most powerful AI model" it has ever developed, the company maintains that limiting access is necessary to prevent catastrophic hacks and terror attacks. The situation marks a pivotal moment in the intersection of artificial intelligence, national security, and financial stability, with government officials and industry leaders racing to establish protocols for managing the technology's dual-use potential.