Topic overview
In brief
- Anthropic's Claude Mythos Preview developed new cryptanalytic techniques against AES-128 and HAWK.
- The research indicates a significant speed increase in recovering encryption keys, but does not compromise the full AES-128 algorithm.
- Experts believe AI could enhance the evaluation of cryptographic systems, but caution against overestimating its capabilities.
Summary
In a recent blog post published on July 28, 2026, representatives from Anthropic's Frontier Red Team announced that their AI model, Claude Mythos Preview, has independently developed new cryptanalytic techniques targeting two significant cryptographic systems: a simplified version of AES-128 and HAWK, an experimental post-quantum digital signature scheme. This research is part of ongoing efforts by the U.S. National Institute of Standards and Technology (NIST) to standardize cryptography for the quantum computing era. The study, authored by scientists Milad Nasr and Nicholas Carlini, has not yet undergone peer review, but it has already garnered attention for its implications on the future of cryptography.
The AES-128 algorithm is widely used to protect various aspects of digital life, including online banking, messaging apps, and Wi-Fi networks. While the full AES-128 algorithm employs 10 rounds of encryption, Anthropic's research focused on a seven-round version, where Claude was able to recover the encryption key significantly faster—between 200 and 800 times quicker than previous methods. However, it is crucial to note that this technique does not apply to the full version of AES-128 used in real-world applications, which remains secure.