The Mythos artificial intelligence model developed by Anthropic identified vulnerabilities in highly sensitive and classified US government computer systems during recent test exercises. According to American officials, the algorithm required only a few hours to detect these security breaches, whereas this process typically takes weeks under normal conditions.
Test Details and the Project Glasswing Initiative
An anonymous US official told the Associated Press that the trials were conducted in collaboration with American intelligence agencies. However, the official emphasized a critical technical distinction: the mere fact that vulnerabilities were discovered did not mean the AI model would have been able to successfully exploit them or carry out a full-scale breach within the same timeframe.
Testing took place under an Anthropic initiative called Project Glasswing. In this project, technology companies join forces to protect critical software. Their primary objective is to prevent the severe consequences for national security and the global economy that experts fear could follow the uncontrolled use of powerful models like Mythos.
Reaction from Congress and the Pentagon
Details of the successful and strikingly fast trials came to light thanks to Virginia Democratic Senator Mark Warner. Speaking at a Senate Banking Committee hearing on June 11, he referenced data from General Joshua Rudd, the head of the National Security Agency (NSA) and US Cyber Command.
"This tool breached nearly all of our classified systems in hours rather than weeks," Warner stated, highlighting the neural network's unprecedented speed in analyzing protected environments.
Growing Tension Between Anthropic and the White House
Despite active collaboration with US intelligence on cybersecurity, ideological tensions are growing between Anthropic and the Donald Trump administration.
Anthropic has openly expressed concerns regarding the specific ways the Pentagon and military structures plan to utilize their AI systems. In response, the administration is taking aggressive steps to restrict the application of these advanced models. In early June, the White House issued a directive ordering Anthropic to block access to its latest developments—the Fable 5 and Mythos 5 models—for foreign nationals. The company itself considers such restrictive measures to be excessive, a stance that only intensifies the conflict surrounding the future of artificial intelligence regulation in the United States.
