The AI model Mythos by Anthropic hacked into secret U.S. systems in just a few hours: a new twist in the cyber race
An event has occurred in the world of cybersecurity that overturns perceptions of artificial intelligence's capabilities. The powerful AI model Mythos, developed by the company Anthropic, managed to identify critical vulnerabilities in highly secure computer systems of the U.S. government during secret exercises. What took entire teams of experts weeks to accomplish, Mythos completed in just a few hours.
Project Glasswing: A Test Drive for the AI Hacker
The tests were conducted as part of an Anthropic initiative codenamed Project Glasswing. The project's goal was to unite technology giants to protect critical global software from threats posed by advanced AI models. During testing, Mythos was connected to U.S. intelligence agencies. The result exceeded all expectations: the model not only found "holes" in the systems but effectively "infiltrated" nearly all of the country's secret networks in hours, not weeks.
Senator Mark Warner, speaking at a hearing before the Banking Committee, confirmed this information, citing data from the head of the National Security Agency and U.S. Cyber Command, General Joshua Redd. This statement was the first public acknowledgment that a new generation of AI is capable of autonomously hacking government systems.
Political Tension Amidst a Technological Breakthrough
Despite successful cooperation in vulnerability discovery, relations between Anthropic and the Donald Trump administration are heating up. The administration issued a directive requiring the company to restrict foreign citizens' access to its latest models — Fable 5 and Mythos 5. Anthropic, in turn, complied with the order by disabling the models for all clients but publicly stated that it considers the authorities' actions unjustified.
Over 100 experts and executives from companies such as Adobe and Nvidia sent a letter to the authorities demanding the directive be revoked. They emphasize that Mythos is a powerful but not unique tool for finding vulnerabilities. Many specialists actively use other basic and open-source models for security auditing. Depriving the country of the best means of cyber defense at a time when U.S. adversaries are rapidly building their capabilities is, in their view, a dangerous and shortsighted strategy.
Cryptalist Analyst's Comment: This case clearly demonstrates the dual nature of advanced AI. On one hand, Mythos shows incredible effectiveness in protecting national interests, acting as a "digital super-scanner." On the other hand, this same tool becomes a point of contention between business and the state. The market clearly sees: control over AI models capable of automated hacking will become a key asset in the new Cold War. Investors should closely monitor regulatory changes in this area, as they will directly impact the value of developers of such technologies.