AI found a weakness in one of the world's most studied encryption systems — is your data under threat?

Recent research conducted by artificial intelligence (AI) has exposed new vulnerabilities in two encryption systems, potentially marking the onset of a new era in digital security. Anthropic's Frontier Red Team revealed that their AI model, Claude Mythos Preview, independently identified new cryptographic techniques against two targets: a version of AES-128, a widely used encryption algorithm, and HAWK, an experimental post-quantum digital signature scheme.

While these findings may not immediately jeopardize individual passwords or bank accounts, they demonstrate AI's potential in uncovering flaws in the mathematical foundations of digital security.

AES, or Advanced Encryption Standard, protects a vast amount of digital data, including websites, messaging apps, Wi-Fi networks, and financial transactions. However, the version Anthropic examined was a seven-round variant of AES, a weakened version employed for testing new attack techniques. The researchers discovered a method to recover the encryption key from this simplified version, making the best-known attack between 200 and 800 times faster.

Nevertheless, this technique does not apply to the full AES-128 algorithm used in real-world systems.

HAWK, on the other hand, is a relatively new digital signature scheme designed to resist attacks from future quantum computers. It is currently under evaluation by the U.S. National Institute of Standards and Technology (NIST) for standardization in post-quantum cryptography. Anthropic's AI, Claude Mythos Preview, found a mathematical property that researchers had previously overlooked, significantly simplifying the recovery of HAWK's secret key when combined with existing attack techniques.

This breakthrough illustrates how AI can accelerate cryptanalysis, enabling researchers to evaluate the strength of cryptographic systems more efficiently before deployment.

Experts emphasize that the threat is not as dramatic as it may seem. The encryption protecting everyday digital life has not been compromised. However, the significance lies in AI's potential to apply expert-level cryptanalysis at scale across thousands of algorithms that researchers have not had time to examine. The HAWK breakthrough, though not solely attributed to AI, showcases the tool's ability to fortify the testing process for future encryption schemes.