What are the implications of the attacks against MD5?
The MD5 attack was first detailed in [Dob96a]. The attack does not currently allow an adversary to create a slightly altered message given an arbitrary message that will match hash values. That is to say given a message A, an adversary cannot feasibly find a message B that hashes to the same hash M. What it does potentially allow an adversary to do is create a message A with a related, but different message B, that hashes to the same M. Practically, you should sign message only when a third party does not have influence over the message being signed. The long and the short of it: MD5 should no longer be used universally without forethought. Users have to be careful when considering which documents to notarise or sign with MD5. Compare this with SHA-1 & RIPEMD with which no such forethought is necessary (because no B can be found that hashes to the same M with these two alternative algorithms). If you are interested then I recommend [Dob96b] for a (slightly outdated) description of the
