Anthropic's Claude Mythos Preview has achieved a significant breakthrough in cryptanalysis, speeding up an attack on seven-round AES-128 by 200 to 800 times and recovering HAWK-256 signing material. Neither result impacts production systems, but the findings highlight the model's advanced capabilities.
TL;DR
- Anthropic's Claude Mythos Preview speeds up AES-128 attack by 200 to 800 times.
- Claude Mythos Preview recovers HAWK-256 signing material, but neither result affects production systems.
What happened
Anthropic's Claude Mythos Preview derived an end-to-end key-recovery attack against HAWK-256 and sped up a seven-round AES-128 attack by 200 to 800 times. The HAWK attack exploits a previously unused symmetry in the lattice behind the signature scheme. The AES result removes a 256-way guessing step from an existing meet-in-the-middle attack.
Anthropic published the findings alongside two technical papers and reproducibility artifacts. The company said Mythos Preview largely conducted the research itself, with humans supplying project direction, computing resources, and extensive verification.
The HAWK attack targets HAWK-256, a challenge parameter provided as a cryptanalytic target. The AES attack assumes an adversary can obtain about 2^105 chosen plaintexts encrypted under one fixed, unknown key, making it impractical for real-world use.
Why it matters
The breakthroughs demonstrate the advanced capabilities of Anthropic's Claude Mythos Preview in cryptanalysis, which could have implications for the development of more secure cryptographic systems. The findings also highlight the potential for AI models to assist in cryptographic research.
For developers and startups, the results underscore the importance of staying updated with the latest advancements in cryptography and AI. Investors may see this as a testament to Anthropic's innovative capabilities, potentially influencing their investment decisions.
However, the practical impact of these findings is limited, as neither the HAWK nor the AES results affect production systems. The attacks remain impractical for real-world use, and the findings are more relevant for academic and research purposes.
Key facts
- Anthropic's Claude Mythos Preview sped up a seven-round AES-128 attack by 200 to 800 times.
- The HAWK attack exploits a previously unused symmetry in the lattice behind the signature scheme.
- The AES result removes a 256-way guessing step from an existing meet-in-the-middle attack.
- Anthropic published the findings alongside two technical papers and reproducibility artifacts.
- The HAWK attack targets HAWK-256, a challenge parameter provided as a cryptanalytic target.
- The AES attack assumes an adversary can obtain about 2^105 chosen plaintexts encrypted under one fixed, unknown key.
- Anthropic estimates that the expected HAWK-256 key-recovery work factor falls from 2^64 to 2^38.
- The model run cost approximately $100,000 in API usage, but researchers spent several hundred hours checking the method.
Context
Anthropic is a leading AI company known for its advanced language models, including Claude. The company has been actively involved in cryptographic research, aiming to enhance the security of digital systems.
The National Institute of Standards and Technology (NIST) is in the process of standardizing post-quantum cryptographic algorithms. HAWK is one of the candidates in this process, and the recent findings could influence its standing.
The development of more secure cryptographic systems is crucial in the face of advancing quantum computing technologies. AI models like Claude Mythos Preview can play a significant role in this endeavor by assisting in cryptanalysis and the development of new cryptographic techniques.
