The Mythos AI model by Anthropic: US cyber shield breached in hours — Cryptalist analysis
An event has occurred in the world of cybersecurity that overturns perceptions of the speed and effectiveness of auditing secure systems. The Mythos artificial intelligence model, developed by Anthropic, has demonstrated alarming efficiency: it identified critical vulnerabilities in classified US government computer networks in just a few hours. This is not a matter of weeks or months of painstaking work by a team of ethical hackers, but a matter of mere hours of algorithm operation.
The testing took place as part of an Anthropic initiative codenamed Project Glasswing. This project brought together technology giants and intelligence agencies with the goal of protecting critical global software. It has become known that Mythos managed to penetrate nearly all of the country's secret systems. However, according to statements from government officials, the model was unable to fully exploit the breaches it found—but the very fact of detection speaks to the technology's enormous potential.
Political Context and Growing Tension
Significantly, information about the testing surfaced during Senate hearings on June 11. Senator Mark Warner, citing data from NSA and US Cyber Command head General Joshua Rudd, confirmed that the tool "penetrated" the systems within hours. This statement was later confirmed by a high-ranking military source. Notably, the NSA and Anthropic itself declined official comment, which only fuels interest in this story.
However, behind this technological triumph lies a serious political conflict. The Trump administration, fearing technology leaks, issued a directive banning the use of Anthropic's latest models—Fable 5 and Mythos 5—by foreign nationals. Furthermore, a risk review mechanism was introduced one month before the public release of any advanced AI system. Anthropic, in turn, was forced to disable the models for all clients to comply with the order, calling the administration's actions unjustified.
More than 100 cybersecurity experts, including representatives from Adobe and Nvidia, sent a letter of protest. They argue that restricting access to Mythos is a "dangerous deprivation of the country's best cyber defense tools" at a time when US adversaries are rapidly building up their capabilities. In their view, Mythos is good at finding vulnerabilities but is not unique—many companies use open models for auditing, and the ban will only weaken the country's defense capabilities.
Expert Commentary: This situation is a classic example of the security and control dilemma. On one hand, we see how AI is changing the rules of the game in cyber warfare, reducing vulnerability detection time from weeks to hours. On the other hand, attempts to restrict this technology out of fear of its misuse could backfire, depriving the defense sector of its most powerful tool. The market makes it clear: regulation here must be nuanced, otherwise we risk ending up with a weakened, rather than protected, infrastructure.