The record
Written from the 1 report below. Nothing here is unsourced.
- Anthropic says its Claude Mythos Preview AI model helped develop a working key-recovery attack against HAWK-256, a cryptanalytic challenge version of a post-quantum signature scheme, and improved an attack on seven-round AES-128 by a factor of 200 to 800.
- The HAWK attack exploits a symmetry in the lattice underlying the scheme, and the finding contributed to HAWK's withdrawal from NIST's post-quantum standardization process.
- The company stressed that neither result affects production systems: the HAWK attack targets only the smaller HAWK-256 parameter rather than the NIST security levels, and the AES attack remains impractical, requiring an enormous number of chosen plaintexts.
- Anthropic published two technical papers and reproducibility code, saying the model did most of the research itself over about 60 hours at an API cost of roughly $100,000.
What to watch next
- Whether NIST or the HAWK submitters formally change the scheme's parameters, security claims, or standing in the standardization process.
- Whether independent researchers verify the released HAWK-256 attack code and the AES speedup estimates.
- Whether the AES technique, which removes a 256-way guessing step, can be extended toward more rounds of AES-128.
Coverage1
1 report
English national1
All filed from India
Named United States · AES · HAWK · Anthropic · Claude Mythos Preview · Daniël van Gent · ETH Zurich · Ludo Pulles · National Institute of Standards and Technology · Technische Universität Berlin · Tel Aviv University · University of Haifa
The 1 report is listed beside the record.
Ask this story
Answers cite the reports above, or say they can't.