Pentagon General: Advanced AI Model Hacked US Cyber Command Systems
The implementation of artificial intelligence is yielding interesting side effects. General Joshua Rudd, head of the US National Security Agency and the Pentagon's Cyber Command, reported that Anthropic's advanced Mythos AI model was able to penetrate virtually all of the agency's classified systems in just a few hours.
The information was published by The Economist.
According to sources, this involved controlled testing, during which Mythos demonstrated exceptional capabilities in finding and exploiting vulnerabilities—and subsequently hacking the system. The model successfully penetrated the defenses of most tested, classified NSA and Cyber Command systems.
In response to the growing risks, the White House has imposed strict restrictions. Access to the latest Mythos 5 and Fable 5 models, as previously reported by VO, is prohibited to foreigners, including citizens of the United States' closest allies in the Anglo-Saxon Five Eyes alliance (the United Kingdom, Canada, Australia, and New Zealand). The restrictions were imposed on short notice and, according to some reports, are related to both national security and political disagreements between the administration and Anthropic.
Mythos is positioned as a specialized, high-powered cybersecurity model. Anthropic previously launched Project Glasswing, a program that provides limited access to the model to trusted defenders for vulnerability scanning. However, the demonstration of its power has raised serious concerns within the government.
Experts note that the incident highlights the dual nature of AI: on the one hand, it's a powerful tool for strengthening cyberdefenses, but on the other, it poses a potential threat if the model falls into the wrong hands. The events surrounding Mythos and Fable 5 have become one of the most striking examples of tensions between tech giants and US authorities in the era of the AI arms race.
- Alexey Volodin
Source: https://en.topwar.ru