, The Register 2002-09-16
Theoretical attacks against AES (Advanced Encryption Standard) winner Rijndael and runner-up Serpent have been published. They might work in the practical world; they might not. That's about all we can say from the latest edition of Bruce Schneier's CryptoGram newsletter, which seeks to simplify the issues discovered by researchers Nicolas Courtois and Josef Pieprzyk, and elaborated in a paper entitled "Cryptanalysis of Block Ciphers with Overdefined Systems of Equations".Now while this represents an interesting bit of research, it does not mean that AES has been or even can be cracked in the real world. The work is theoretical and needs to be reviewed by others; and even if it's confirmed in theory and partially confirmed empirically, it may never be possible to exploit it.
"You can try the attack on simplified versions of the cipher -- fewer rounds, smaller block size -- but you can never be sure the attack scales as predicted," Schneier points out.
That said, it's ironic that Serpent, which touts itself as more secure than Rijndael (though slower), appears at least for now to be more vulnerable in this case. And while there's no immediate problem for either cipher, we may find that AES' functional life-expectancy is considerably shorter than originally hoped (something like a century).
"If the attack really works, it can only get better. My fear is that we could see optimizations of the XSL attack breaking AES with a 2^80-ish complexity, in which case things starts to get dicey about ten years from now," Schneier reckons.
That's bad news for those who might require that an encrypted object remain impractical to decrypt for several decades. Some businesses perhaps, and certainly some government and military agencies do need very long-lasting and very strong levels of communications and data security. What we're seeing here is another example proving that asking for one-shot solutions is asking too much. We've said it before and we'll say it again: there's no substitute for a holistic approach to security and privacy in which no single component is ever fully trusted. ®