In order to defend against adversaries with undisclosed capabilities, you have to extrapolate known attack methods and hardware to produce estimates of what may be practical. Every paper that proposes keylength and parameter size recommendations does this. If we didn't extrapolate this way, there would have been no reason to stop using 56-bit symmetric ciphers until 1998 (!). The hardware necessary to crack such a cipher could have been dismissed as "magical" because nobody who had built it had published a paper about it.