The AI model Mythos by Anthropic hacked secret US networks in hours: a new era of cyber warfare
The world of cybersecurity has experienced a tectonic shift. The Mythos artificial intelligence model, developed by Anthropic, demonstrated frightening effectiveness by penetrating the highly secure computer systems of the U.S. government within hours. This is not about weeks or months of work by entire teams of hackers, but time measured in hours.
Project Glasswing: a secret alliance of technology and intelligence
Under the codename Project Glasswing, Anthropic joined forces with U.S. intelligence agencies. The goal of the testing was to assess Mythos's ability to find critical vulnerabilities in software of strategic importance to national security and the economy. The results were staggering: the AI identified numerous vulnerabilities in a matter of hours, although sources emphasize this does not mean the model could have immediately exploited them.
Democratic Senator Mark Warner of Virginia confirmed this information at a hearing on June 11, stating that the tool had "penetrated" virtually all of the country's secret systems. His words were backed by a high-ranking military source. Notably, the National Security Agency (NSA) and representatives of Anthropic declined official comment, which only fuels interest in the details of this breakthrough.
Geopolitical tension and AI regulation
This collaboration occurs against a backdrop of growing tension between Anthropic and the administration of Donald Trump. Earlier this month, the administration issued a directive banning foreign nationals from using Anthropic's latest models — Fable 5 and Mythos 5. This decision came just ten days after the president's executive order to establish a risk review mechanism for advanced AI systems.
Anthropic, for its part, expressed concern about how the U.S. military would use its technologies and stated its disagreement with the authorities' actions. Over 100 experts and executives from companies including Adobe and Nvidia sent a letter requesting the directive be rescinded, warning that it "deprives the country of the best cyber defense tools without good reason at a time when U.S. adversaries are rapidly building capabilities."
Cryptalist analytical conclusion: This case is not just a demonstration of AI's power in pentesting. It is a direct signal that the arms race in cybersecurity is moving to a new, autonomous level. While regulators try to restrict access to such technologies, real threats and opportunities are already being tested in "combat" conditions. For the crypto industry, where the security of smart contracts and infrastructure is critical, this means AI-based audit tools will become not just an advantage, but a necessity. The question is not whether such systems will be used, but who will have access to them first.