Measuring LLMs’ Ability to Perform Cryptanalysis

A new benchmark, CryptanalysisBench, measures AI's ability to perform mathematical cryptanalysis. Anthropic's frontier model found new attacks on Hawk and reduced-round AES. This is a concern for digital security, as AI's ability to break cryptographic schemes may soon match or exceed the published state of the art. Engineers should be aware of this development and monitor its impact. Further testing and evaluation are needed to fully understand the implications.