A hash algorithm turns an ordinary message into a "digital fingerprint," which can then be used to keep the original message secret during transit or to guarantee that it hasn't been tampered with en route. But a hash function is only considered secure if there is no practical way to run it backward and find the original message from the fingerprint. Equally important, there should be no trivial way to produce two messages with exactly the same fingerprint. The weaknesses discovered by Wang and others relate to this problem--something cryptographers call "a collision." The latter issue is complicated by the fact that it is impossible to completely avoid collisions. So the best algorithm is one that simply makes collisions extremely hard to produce. "You shouldn't be able to find them," says William Burr, manager of the Security Technology Group for NIST. "The computation should be too great."