Mythos model found vulnerabilities in classified US Government systems (apnews.com)

Anthropic's Mythos model has uncovered vulnerabilities in classified U.S. government systems during a collaborative testing initiative with U.S. intelligence agencies. This revelation highlights the model's capability to identify security weaknesses in sensitive environments within hours, although exploiting them was not achieved. The testing was part of Project Glasswing, aimed at safeguarding critical software in light of potential risks that advanced AI models like Mythos might pose to national security and public safety. The findings have sparked discussions about the implications for AI's role in cybersecurity and its oversight. The incident is significant as it underscores the dual-edged nature of AI technologies—capable of both aiding in the identification of vulnerabilities and raising concerns about their possible misuse. Tensions have arisen between Anthropic and the U.S. government, particularly after an executive order restricted the use of advanced AI models like Mythos and Fable 5, mandating safeguards against foreign access. Despite industry pushback from cybersecurity leaders questioning the restrictions, the broader discourse on AI’s security implications continues, emphasizing the need for balanced access to these powerful tools amidst growing cyber threats from adversaries.