Anthropic's Claude Cuts HAWK Key Strength by Half, Accelerates AES Attack 200 to 800 Times
Anthropic revealed that Claude, using its Mythos Preview model, identified weaknesses in two cryptographic algorithms, finding a vulnerability in the proposed HAWK digital signature scheme that halves the key strength and discovering a method to accelerate an attack on a reduced version of the Advanced Encryption Standard (AES) by 200 to 800 times.
The research was conducted autonomously with occasional human guidance, with each result costing approximately $100,000 in API usage. Anthropic disclosed the findings to the algorithms' authors and U.S. government partners before publication and emphasized the results have no practical impact on current systems, as HAWK has not been deployed and the AES analysis targets a reduced cipher rather than the full standard. Developed in collaboration with academics from ETH Zurich, Tel Aviv University, and the University of Haifa, the work demonstrates frontier AI's ability to perform expert-level cryptography research with potential defensive applications for testing security algorithms.
From the sources (11 posts)
@anthropicaiNew Anthropic research: Discovering cryptographic weaknesses with Claude. Claude Mythos Preview has helped our researchers find weaknesses in cryptographic algorithms—the mathematical methods that are used to keep data private. Read more:
@anthropicaiClaude discovered weaknesses in a highly-secure digital signature scheme (used to verify identity digitally) and a well-known symmetric cipher (used to encrypt data).
@anthropicaiThe digital signature scheme is HAWK, which is designed to be robust even against hypothetical quantum computers. HAWK has survived two years of expert review, but in 60 hours Mythos Preview found a previously-unknown attack that reduced t
@anthropicaiThe symmetric cipher is a reduced version of the Advanced Encryption Standard (AES)—which has received decades of scrutiny (more than almost any other encryption algorithm). In a week, Mythos Preview found a way to speed up an attack on th
@anthropicaiMythos Preview did most of this work autonomously, with occasional human guidance. Each of the two results cost roughly $100,000 in API usage. We disclosed the findings in advance to the algorithms’ authors, as well as to US government and
@anthropicaiThese are substantial research advances, but they don’t have a practical impact on today’s systems. HAWK is a proposed scheme that hasn’t been deployed anywhere, and the AES attack we discovered was on a weaker version and does not break th
@anthropicaiStill, both results show that frontier AI models are capable of doing expert-level cryptography research. This has important defensive applications—testing the algorithms that keep our online activity secure, and ultimately helping to make
@anthropicaiFull technical details of both attacks are provided in our new papers: On HAWK: On AES: And the associated model chain-of-thought for AES:
@anthropicaiWe also worked with academics at ETH Zurich, Tel Aviv University, and the University of Haifa to build CryptanalysisBench, a benchmark for studying LLMs’ cryptanalysis abilities.
@degeneratenewsNEW: @AnthropicAI ANNOUNCES THAT RESEARCHERS, USING CLAUDE MYTHOS PREVIEW, FOUND IMPROVED WAYS TO ATTACK CRYPTOGRAPHIC ALGORITHMS - THE MATHEMATICAL METHODS THAT SAFEGUARD ONLINE DATA PRIVACY ONE ATTACK SIGNIFICANTLY WEAKENS HAWK (A POST-Q
@cointelegraph🚨 BIG: Anthropic’s Claude helped researchers discover weaknesses in a highly secure digital signature scheme and a well-known symmetric cipher used to encrypt data.