Anthropic Uncovers New Cryptographic Attacks with Claude

In a groundbreaking revelation, Anthropic announced that its advanced AI model, Claude Mythos Preview, has identified novel ways to attack two cryptographic algorithms: the post-quantum digital signature scheme HAWK and a simplified version of the AES cipher. These discoveries underscore the transformative potential of modern AI models in analyzing encryption methods while reassuring that they do not endanger current systems.

Claude Finds a Weakness in HAWK Algorithm

The first significant finding involves HAWK, one of the digital signature schemes participating in the National Institute of Standards and Technology (NIST) competition for selecting post-quantum cryptographic algorithms. According to Anthropic’s report, Claude took approximately 60 hours to identify a new attack method that effectively halves the cryptographic strength of HAWK keys. The AI model discovered a mathematical feature within the algorithm’s structure that accelerates key recovery significantly.
Despite these findings, it’s important to note that HAWK is still undergoing standardization and is not yet deployed in real-world systems. This proactive approach ensures these algorithms are openly scrutinized for potential weaknesses before implementation. Anthropic has responsibly shared its findings with HAWK authors and coordinated information disclosure with NIST.

AI Enhances Attack on Simplified AES Version

The second breakthrough concerns AES, one of the most widely used symmetric encryption algorithms. In this case, Claude identified a more effective way to attack an experimental version of AES with seven rounds instead of the standard ten. These shortened versions are traditionally employed in cryptographic research to assess an algorithm’s resilience.
Anthropic estimates this new method operates 200-800 times faster than previous similar attacks. However, it is crucial to emphasize that this discovery does not compromise the full version of AES used in contemporary products.

The Broader Impact on Cryptography

Beyond these two primary investigations, Claude also found new attacks on simplified versions of LEA and Serpent algorithms while enhancing certain analysis methods for Salsa20, Poseidon, and SHA-1. To further assess language models’ capabilities in cryptanalysis, Anthropic introduced CryptanalysisBench as a benchmark tool.
Analysts anticipate that AI will become an invaluable tool for vetting cryptographic algorithms before deployment. However, as models’ capabilities expand, there will be an increasing need for new approaches to responsibly disclose potentially critical cryptographic vulnerabilities.
This remarkable development by Anthropic represents just one instance where artificial intelligence is reshaping our understanding and application of security technologies within the cryptocurrency landscape. As we move forward into an era where digital security remains paramount, such advancements provide both challenges and opportunities for ensuring robust protection against emerging threats—a dynamic which will undoubtedly continue shaping future discourse around cryptocurrency security practices.