Anthropic AI Finds Major Flaws in Encryption Systems

Artificial intelligence just took a major step forward in cybersecurity. Anthropic revealed that its Claude model successfully found new weak spots in two major encryption systems. Encryption is the crucial math that keeps messages, logins, and financial data hidden from bad actors. While no current user data is at risk right now, a machine just did work that top human cryptographers missed for years.

The AI targeted HAWK, which is a digital signature built to survive future quantum computers. After reviewing the system for two years, human experts found zero issues. Claude spotted a hidden mathematical pattern in just sixty hours. The AI reduced the work needed to crack the system by millions of times. The fix requires doubling the key size, which takes away the main advantage of using HAWK in the first place.

Claude also targeted a reduced version of AES, the standard protocol that has guarded internet traffic since 2001. The model invented a new shortcut that made the best known attack hundreds of times faster. Anthropic notes that checking the work is now the slow part because humans cannot keep pace with the speed of the model.

Bitcoin and Ethereum rely on different systems, and these specific attacks do not threaten them today. Even so, the clock is running as major tech companies and government agencies prepare for future quantum deadlines. Watch for upcoming announcements from NIST and academic workshops as researchers figure out how to keep up with AI security testing.