A recent viral claim that "AI broke into the NSA in hours" has been clarified as a misunderstanding. The event was actually an authorized red team exercise, a standard security practice where teams simulate attacker behavior to test defenses in a controlled environment. During the exercise, an AI model named Mythos efficiently identified and exploited a series of vulnerabilities, demonstrating the advanced potential of cutting-edge AI in cybersecurity.
The performance of the Mythos model has sparked deep discussions about the security capabilities of AI. On one hand, AI's ability to quickly find vulnerabilities can enhance system security; on the other hand, its powerful offensive capabilities raise regulatory concerns. How to balance technological innovation and security risks has become a key focus for stakeholders.
Meanwhile, Anthropic faces U.S. export controls due to risks associated with its AI model. However, it is noteworthy that agencies like the National Security Agency (NSA) continue to use Anthropic's preview version internally. This situation highlights a contradiction between policy and practice: export restrictions are imposed while the same technology is relied upon domestically.