US Treasury Secretary Praises Anthropic's Mythos AI as Crucial in Race Against China
In a significant development, AI giant Anthropic PBC has received a major endorsement from the Trump administration after a contentious six-week legal fight with the Pentagon. The reason behind this support is the intensifying competition with China in the artificial intelligence sector.
High-Level Praise for Anthropic's Revolutionary Model
Speaking at a Wall Street Journal event in Washington, US Treasury Secretary Scott Bessent lauded Anthropic's new Mythos model, describing it as a revolutionary step that will help America maintain its technological edge over China. Bessent emphasized the model's advanced capabilities, stating, "This Anthropic Mythos model was a step function change in abilities, learning capabilities. It's all logarithmic. You go from x to the 10th power to x to the 12th and then it's very difficult to catch up."
While acknowledging that the US advantage might be as narrow as three to six months, Bessent dismissed suggestions that China is rapidly closing the gap. This praise marks a stark contrast to earlier this year when the Pentagon labeled Anthropic as a threat to the US supply chain, invoking powers typically used for foreign adversaries.
Legal Victory and Cybersecurity Concerns
Anthropic successfully fought back in court, securing an order that blocked a government ban on the use of its technology—a move the company warned could have cost billions in lost revenue. The focus now shifts to Anthropic's latest AI model, Mythos, which has raised significant cybersecurity concerns due to its ability to find vulnerabilities in software and computer systems.
Mythos is being released to a limited number of carefully selected parties, a decision that has sparked debates about potential cyber risks. Just days before Bessent's remarks, he and Federal Reserve Chair Jerome Powell convened Wall Street banks to discuss these very risks, highlighting the model's potential to exploit software vulnerabilities more effectively than nearly any human.
Uncovering Hidden Flaws and Controlled Release
Anthropic reports that the Mythos model has already uncovered thousands of severe, previously unknown flaws across every major operating system and web browser. For instance, one vulnerability in OpenBSD—widely regarded as one of the most secure operating systems—had gone undetected for 27 years. Another flaw in the widely used video tool FFmpeg was located in a line of code that automated testing tools had hit five million times without identifying the issue.
To manage these risks, Anthropic is limiting access to select partners, including Google, Microsoft, JPMorgan Chase, and CrowdStrike, under a program called Project Glasswing. This initiative aims to harness Mythos-class capabilities for defensive purposes in a controlled environment, reflecting the company's reputation as a "safety-first" AI firm.
Anthropic has emphasized that an uncontrolled release of such powerful AI could have severe consequences for economies, public safety, and national security, underscoring the delicate balance between innovation and risk management in the AI race.
