In a high-stakes security exercise under the restricted "Project Glasswing" initiative, Anthropic’s Mythos AI model successfully identified significant vulnerabilities within sensitive US government computer systems. According to reports, Senator Mark Warner revealed during a recent congressional hearing that NSA chief Joshua Rudd confirmed the model managed to navigate and uncover weaknesses in nearly all classified networks in a matter of hours. While officials noted that detecting a vulnerability is distinct from successfully exploiting one, the rapid pace of the discovery highlights the transformative and potentially disruptive power of modern AI in the cybersecurity landscape.

The relationship between Anthropic and the US government remains increasingly tense, marked by significant policy friction. While the company has taken a firm stance against integrating its technology into domestic surveillance or autonomous weapons systems, this ethical boundary has led to retaliatory measures, including the model’s placement on a national security blacklist. Recently, the situation escalated further when the government restricted the export of Anthropic’s advanced AI models, Mythos and Fable, to international markets and foreign nationals. As a result of these ongoing disputes, the NSA has reportedly lost access to the Mythos system, underscoring a deepening divide between private AI innovation and federal security mandates.