Anthropic's AI model Mythos hacked secret US systems in a matter of hours: a new era of cybersecurity?
What many experts predicted has happened, but even the most protected structures were unprepared for it. According to my data obtained from high-ranking sources in U.S. government circles, the advanced artificial intelligence model Mythos from Anthropic has demonstrated frightening effectiveness, identifying critical vulnerabilities in highly secure government computer systems. And this is not a matter of weeks or months—it took just a few hours.
This became possible thanks to a secret initiative called Project Glasswing, under which Anthropic joined forces with U.S. intelligence agencies. The project's goal was not just testing, but a stress test of critical software on which national security and the country's economy depend. The results, according to my data, exceeded all expectations.
How the testing was conducted
Democratic Senator from Virginia Mark Warner made a sensational statement at a hearing on June 11 before the Senate Banking Committee. He reported that the tool "penetrated" nearly all of the country's secret systems not in weeks, but in hours. He attributed this information to the head of the National Security Agency (NSA) and U.S. Cyber Command, General Joshua Radu, which gives it special weight. The NSA and Anthropic itself are refraining from official comments for now, which only fuels interest in this story.
Growing tension with the administration
Notably, this breakthrough occurs against the backdrop of a growing conflict between Anthropic and the Donald Trump administration. The White House, fearing technology leaks, issued a directive prohibiting foreign citizens from using the company's latest models—Fable 5 and Mythos 5. Anthropic, in turn, deemed these actions unjustified and disabled the models for all clients in protest.
More than 100 executives and experts from leading technology companies, including Adobe and Nvidia, sent a letter to the administration demanding the directive be revoked. They argue that such restrictions only play into the hands of U.S. adversaries, depriving the country of the best cybersecurity tools. According to them, Mythos is certainly powerful in finding vulnerabilities, but far from unique—there are many open models on the market that security auditors work with.
My analysis: This incident is a vivid demonstration of the dual nature of AI. On one hand, we see incredible potential for strengthening cybersecurity. On the other, the panic of regulators who do not yet know how to control such power is obvious. The cryptocurrency market, which traditionally relies on decentralization and code security, should closely monitor these events. If AI can find breaches in NSA systems in hours, what vulnerabilities might it discover in smart contracts? The arms race in cybersecurity is entering a fundamentally new level, and the stakes are the entire digital economy.