A U.S. official disclosed that Anthropic's AI model, Mythos, identified vulnerabilities in secure U.S. government systems during testing with intelligence agencies. This effort is part of Project Glasswing, aiming to address significant public safety and national security concerns. Although Mythos found vulnerabilities rapidly, it did not exploit them. Tensions have risen between Anthropic and the Trump administration, particularly regarding the military's use of AI.
The administration implemented directives limiting the use of some AI models due to security fears, while cybersecurity executives have urged the administration to lift these restrictions, arguing that such limitations could inadvertently benefit adversaries.