2021. MD5.
I can't wait to hear the apologists.
EDIT: instead of downvoting it would be nice to explain why it's fine to use and encourage the usage of MD5 in 2021.
2021. MD5.
I can't wait to hear the apologists.
EDIT: instead of downvoting it would be nice to explain why it's fine to use and encourage the usage of MD5 in 2021.
1) All of the tools expect and produce MD5 because it is the convention. Computing MD5 hashes of every file on disk or passing through a network is a relatively common forensic operation right now, SHA256 is not.
2) IOCs are not intended to be used in scenarios in which a malicious collision presents a problem (no one would want to... mask their malware to still look like malware?) so there is little downside to carrying on the convention.
While I would recommend against MD5 in most modern applications, if nothing else to avoid having discussions like this all the time, before upsetting an entire ecosystem of tools it is important to consider whether or not the known weaknesses of MD5 actually pose a problem. In this case, a matching hash is the bad state, and so there is no real impact of a preimage attack.
Now, if you used it to verify whether a binary was _secure_, this would be problem. But in this case, the (still unlikely) possibility of a false positive is not really a threat.
Even that's probably fine. Collision attacks require the attacker to control both inputs. In the case of code signing this would mean the publisher is in on it, in which case you're already screwed.
Technically, no. It gets a lot easier when you only need to find any pair of inputs which produce a collision, but with computing power increasing a lot and md5 being very broken, it is feasible that a determined attacker with a lot of compute power can create a collision with a fixed hash, or at least will be able to in the somewhat near future.
I'd agree that you're reasonably secure, but when you need security, you shouldn't bother with outdated tech when secure and future-proof alternatives are readily available.
That's a preimage attack, which is a much higher bar. Even for md5 there hasn't been much progress made: https://en.wikipedia.org/wiki/MD5#Preimage_vulnerability
If the NSA found a way to get the complexity down a bit more (they had a decade) and the target is valuable enough (imagine, for example, having a "valid" TrueCrypt executable), this is not in the realm of impossible anymore.
Now, of course, the NSA will probably not outcompute the Bitcoin blockchain and transfering the hash rate is not that simple, but they do have a lot of power and cryptographers. What I'm trying to say is that the orders of magnitude are not in the "totally impossible" space anymore. Yes, you're probably safe, but given the availability of secure alternatives, there's really no reason not to get off the thin ice.
If MD5 was being used to verify a piece of software you actually want, it's not secure as it's not collision resistant.
But since we can be quite sure no one has made a file that shares an MD5 hash with this new strain of malware, MD5 is sufficient as a checksum in this use case.
You're correct to point out that newer hashes are still preferable though, simply to get out of the habit of using MD5 if nothing else. I assume you got downvoted because MD5's weaknesses aren't relevant in this specific instance. But still they could just have easily used SHA256.
You were downvoted (or, rather, I downvoted you) for your knee jerk "MD5 BAD!" comment without thinking through what the problems would be, and worse, you took an aggressive tone from the beginning ("I can't wait to hear the apologists") that just makes you sound like a jerk.
While your overall point is correct, there is a specific tactic where a collision attack would be useful: make two md5-colliding versions of the virus that behave subtly differently, in a attempt to make researchers throw out useful hypotheses about how it works because someone already determined that (the other version of) it didn't actually do that.
I don't really see how the weaknesses of MD5 are applicable here but I'd like to learn if I'm wrong.