Anthropic’s Mythos model found vulnerabilities in classified US government systems, official says (opens original article in a new tab)

A U.S. official said Anthropic's AI model identified vulnerabilities in classified government systems during testing, leading to government restrictions on the model's use, despite concerns from cybersecurity experts.

• A U.S. official said Anthropic's Mythos model found vulnerabilities in classified government systems during testing
• Testing was part of Project Glasswing aimed at securing critical software from AI risks
• The Trump administration restricted use of Anthropic's models, citing national security concerns
• Cybersecurity experts argued the restrictions could harm U.S. defenses by limiting access to effective AI tools