AI Models Uncover Vulnerabilities in Government Systems

Concerns over cybersecurity are on the rise as AI models have apparently revealed weaknesses in secured government systems. This news comes amidst growing worries among corporations and government officials.

U.S. officials reported that an advanced AI model from Anthropic, known as Mythos, successfully identified flaws in “sensitive and secure U.S. government computer systems during test exercises.” These findings were shared by the Associated Press.

One statement highlighted, “This tool penetrated nearly every sensitive system in hours—not weeks.”

Officials, who wished to remain anonymous, indicated that these tests were intended as part of a broader strategy, particularly related to a program named Project Glasswing. However, it’s important to note that just because a vulnerability has been discovered does not imply it has been exploited.

Project Glasswing was initially announced on April 7. It was described as providing capabilities for defensive measures with the Mythos model.

Furthermore, the project aims to collaborate with industry leaders to find vulnerabilities, committing up to $100 million in credits for Mythos use and an additional $4 million in donations to open-source security organizations. Partners include companies such as Amazon Web Services, Apple, Google, and Microsoft.

Senator Mark Warner (D-Va.) referenced these defenses during a Senate hearing on June 11, emphasizing the speed at which Mythos could identify vulnerabilities, attributing these insights to Gen. Joshua Ladd, who leads the National Security Agency.

In an interesting twist, shortly after a presidential order, which aimed to establish a national security framework for AI and cybersecurity, the government limited access to Anthropic’s latest models for foreign users. This directive led to the temporary suspension of Mythos.

Blaze News attempted to contact both the National Cyber Director and Anthropic for comments, but did not receive a response right away.