Anthropic Mythos Model Flags Vulnerabilities In U.S. Government Systems
Anthropic’s Mythos artificial intelligence model identified vulnerabilities in highly sensitive U.S. government computer systems during a testing exercise, according to a report by the Associated Press on Tuesday.
The testing took place under Project Glasswing, a restricted programme designed to identify and address weaknesses in critical software before attackers could exploit them. Anthropic worked alongside U.S. intelligence agencies during the exercise, with the goal of strengthening the security of government systems.
Project Glasswing Testing Effort
According to the Associated Press, Project Glasswing focuses on uncovering vulnerabilities in essential software used within government infrastructure. The programme aims to detect security flaws early and enable agencies to resolve them before they become targets for malicious actors.
As part of the initiative, Anthropic deployed its Mythos model to assess the resilience of classified systems. The exercise formed part of a broader effort to evaluate how advanced artificial intelligence tools can contribute to cybersecurity and vulnerability detection.
Claims Raised During Congressional Hearing
Senator Mark Warner of Virginia referred to the testing during a congressional hearing earlier this month. He stated that National Security Agency chief Joshua Rudd had informed him that Mythos “broke into almost all of our classified systems, not in weeks, but in hours.”
However, the Associated Press cited an unidentified U.S. official who clarified the scope of the achievement. Although Mythos reportedly identified certain vulnerabilities within hours, the official said this did not mean the model successfully exploited those weaknesses within the same timeframe.
As a result, the distinction between identifying vulnerabilities and exploiting them remains significant in understanding the outcome of the testing exercise.
Tensions Between Anthropic And The U.S. Government
Anthropic’s relationship with the U.S. government has faced challenges as the company prepares for a public offering. The company declined to permit the U.S. military to use its artificial intelligence models for domestic surveillance and fully autonomous weapons systems.
Subsequently, the government placed Anthropic on a national security blacklist. In addition, U.S. authorities ordered the company this month to suspend exports of its latest Mythos and Fable AI models to destinations worldwide and to all foreign nationals, citing national security concerns.
Meanwhile, The New York Times reported earlier on Tuesday that the National Security Agency had lost access to Mythos amid the ongoing dispute.
The White House, Anthropic and the Department of Defense did not immediately respond to Reuters requests for comment.
With inputs from Reuters