I really never managed to understand why we keep suggesting SHA-256 even though we know it's vulnerable to length extension attacks?
Everywhere else in cryptography we favor recommendations that are harder to misuse, or that shrink the surface area of "things you have to think about and get right", but for some reason SHA-256 keeps finding its way into new designs (such as Git 3.0), even though things would be much nicer if I never had to remember that length extensions are a thing :/
On another note, as someone who defaults to BLAKE3 unless I have a good reason not to, it would be nice to have more context on Filippo's comment about it being very fast "but complex and uncommon enough that you want to have a good reason to use it and a good reason to trust your implementation".
I've peeked into the guts of Zig's BLAKE3 implementation before, it's merkle-tree-ish and not very scary looking, the primitives are a lot more recognizable than Keccak for example...
Because length extension attacks only matter if you're doing something you should never be doing: hand-rolling your own keyed hash constructions. Everybody uses HMAC (you even see it in SHA3 systems), and so length extension doesn't matter. Meanwhile, SHA2 has hardware support.
I like BLAKE3 just fine. Nobody's really going to ding you for using it (or SHA3 for that matter). But it would not be sound practice to tear SHA2 out of a system that was already using it.
Two other notes:
(1) The article explicitly recommends HMAC-SHA2, not SHA2 by itself.
(2) Where it does recommend SHA2 directly (for hashing, not authentication), it recommends 512/256, which is truncated and can't be length-extended.
I think the original comment still stands though right? You can hand roll a keyed hash if you use sha3 or a truncated sha2. It saves you having to do two hash operations
You shouldn't do that even with SHA3; SHA3 KMAC isn't literally just secret-prefix keyed SHA3, it has a bunch of encoding things (like commitment to message length). You should essentially never be hand-rolling an authenticator.
Filippo is my favorite way to keep myself abreast of the newest (professionally relevant!) drama in the cryptography community. As a colleague said today, "life is too short to read through all the TLS working group mailing list".
Meh don't care about the actual PQ parts. I'm with Peter Gutmann [0] in that PQ is just a waste of time and draining resources that could be put to better use.
Guttman's talk is entertaining, but his argument around attacker behavior does not hold water: the attacker persona who is willing to do something very noisy (like use SQL injection on a server) is not the same persona that's interested in cryptographic breaks. OWASP measures the former, not the latter.
I didn't really find that the interesting part; more the (lack of) progress in cryptanalysis. I know there's been some brouhaha over the last few months that there's a target date for a CRQC (cryptographically relevant quantum computer) circa 2030. Rather than quibble about how it's feasible or how we should heed the threat, I'd rather just put money on it -- happy to bet anyone interested using the longbets platform.
I don't really have a dog in the CRQC fight. But reasoning abductively: every cryptography I know is at least abstractly concerned about classical breaks, particularly within the last 18 months.
(There's a separate reason to be interested in "PQC" cryptography: even in the absence of a "full" classical break, most cryptographers agree that it's not ideal that all of our eggs are currently in one asymmetric cryptographic "basket," i.e. problems that generalize to the hidden subgroup problem. Asymmetric schemes whose hardness does not depend on a lack of solution to problems in that space seems like a reasonable thing to research and build.)
This gets into the weird history of these "Right Answers" documents, which I deeply regret ever help make into a thing. Colin Percival wrote the first one, in the long-long-ago, and he's the author of Tarsnap, a (very good) encrypted backup service. I wrote the next "Right Answers" document as a response to Percival's, because I felt his "Right Answers" were very idiosyncratic and wanted to put conservative normie answers on the record. Because I like Colin (and Tarsnap), my "Right Answers" had a Tarsnap recommendation, and so have subsequent ones.
I think it's past time for this "Right Answers" cryptography meme to die, though. Filippo's PQC recommendations are good. Let's hope this is the last such document we ever need.
> I think it's past time for this "Right Answers" cryptography meme to die, though.
Politely disagree. I read Colin's when it came out and it was as useful in 2009 as Filippo's is today. If you only pay attention to the cryptography community every few years when you need it for something, then usually a whole lot has happened in the interval. It is incredibly useful to get the perspective of someone paying attention to give you a breakdown of "where we're at today".
A great example is the other conversation about SHA-256. Colin's recommendation was "hey, better plan to migrate from this in the future." The thing that's happened between then and now is, well, SHA-256 is actually still 'ok' when used in current constructions. That's useful to know in 2026.
If I get busy building stuff, it'll be nice to have a check-in again in a few years (or earlier if something significant happens first).
I really never managed to understand why we keep suggesting SHA-256 even though we know it's vulnerable to length extension attacks?
Everywhere else in cryptography we favor recommendations that are harder to misuse, or that shrink the surface area of "things you have to think about and get right", but for some reason SHA-256 keeps finding its way into new designs (such as Git 3.0), even though things would be much nicer if I never had to remember that length extensions are a thing :/
On another note, as someone who defaults to BLAKE3 unless I have a good reason not to, it would be nice to have more context on Filippo's comment about it being very fast "but complex and uncommon enough that you want to have a good reason to use it and a good reason to trust your implementation".
I've peeked into the guts of Zig's BLAKE3 implementation before, it's merkle-tree-ish and not very scary looking, the primitives are a lot more recognizable than Keccak for example...
Because length extension attacks only matter if you're doing something you should never be doing: hand-rolling your own keyed hash constructions. Everybody uses HMAC (you even see it in SHA3 systems), and so length extension doesn't matter. Meanwhile, SHA2 has hardware support.
I like BLAKE3 just fine. Nobody's really going to ding you for using it (or SHA3 for that matter). But it would not be sound practice to tear SHA2 out of a system that was already using it.
Two other notes:
(1) The article explicitly recommends HMAC-SHA2, not SHA2 by itself.
(2) Where it does recommend SHA2 directly (for hashing, not authentication), it recommends 512/256, which is truncated and can't be length-extended.
I think the original comment still stands though right? You can hand roll a keyed hash if you use sha3 or a truncated sha2. It saves you having to do two hash operations
You shouldn't do that even with SHA3; SHA3 KMAC isn't literally just secret-prefix keyed SHA3, it has a bunch of encoding things (like commitment to message length). You should essentially never be hand-rolling an authenticator.
Excellent post as always.
Filippo is my favorite way to keep myself abreast of the newest (professionally relevant!) drama in the cryptography community. As a colleague said today, "life is too short to read through all the TLS working group mailing list".
Would've been interesting to see something about when to use MLKEM-1024-X448 and ML-DSA-87-ed448.
Hurray for mostly-non-controversial opinions!
DJB would like to have a word with you ;) (I'm not sure if this article is not even a semi-troll ^^ - not that especially disagree with it though)
Okay - so... they're saying... don't use p384? :3
As far as I know, P384 has never been a “right answer.”
Meh don't care about the actual PQ parts. I'm with Peter Gutmann [0] in that PQ is just a waste of time and draining resources that could be put to better use.
[0] https://www.youtube.com/watch?v=xa4Ok7WNFHY
Guttman's talk is entertaining, but his argument around attacker behavior does not hold water: the attacker persona who is willing to do something very noisy (like use SQL injection on a server) is not the same persona that's interested in cryptographic breaks. OWASP measures the former, not the latter.
I didn't really find that the interesting part; more the (lack of) progress in cryptanalysis. I know there's been some brouhaha over the last few months that there's a target date for a CRQC (cryptographically relevant quantum computer) circa 2030. Rather than quibble about how it's feasible or how we should heed the threat, I'd rather just put money on it -- happy to bet anyone interested using the longbets platform.
I don't really have a dog in the CRQC fight. But reasoning abductively: every cryptography I know is at least abstractly concerned about classical breaks, particularly within the last 18 months.
(There's a separate reason to be interested in "PQC" cryptography: even in the absence of a "full" classical break, most cryptographers agree that it's not ideal that all of our eggs are currently in one asymmetric cryptographic "basket," i.e. problems that generalize to the hidden subgroup problem. Asymmetric schemes whose hardness does not depend on a lack of solution to problems in that space seems like a reasonable thing to research and build.)
Why can't I use rclone?
This gets into the weird history of these "Right Answers" documents, which I deeply regret ever help make into a thing. Colin Percival wrote the first one, in the long-long-ago, and he's the author of Tarsnap, a (very good) encrypted backup service. I wrote the next "Right Answers" document as a response to Percival's, because I felt his "Right Answers" were very idiosyncratic and wanted to put conservative normie answers on the record. Because I like Colin (and Tarsnap), my "Right Answers" had a Tarsnap recommendation, and so have subsequent ones.
I think it's past time for this "Right Answers" cryptography meme to die, though. Filippo's PQC recommendations are good. Let's hope this is the last such document we ever need.
> I think it's past time for this "Right Answers" cryptography meme to die, though.
Politely disagree. I read Colin's when it came out and it was as useful in 2009 as Filippo's is today. If you only pay attention to the cryptography community every few years when you need it for something, then usually a whole lot has happened in the interval. It is incredibly useful to get the perspective of someone paying attention to give you a breakdown of "where we're at today".
A great example is the other conversation about SHA-256. Colin's recommendation was "hey, better plan to migrate from this in the future." The thing that's happened between then and now is, well, SHA-256 is actually still 'ok' when used in current constructions. That's useful to know in 2026.
If I get busy building stuff, it'll be nice to have a check-in again in a few years (or earlier if something significant happens first).
rclone isn't a backup tool, it's a file transfer tool
Never heard of rclone but probably not a Cryptographic right answer because encryption isn't always on.
> Password handling: Argon2id or scrypt. But have you considered passkeys instead?
I hate passkeys and if your service doesn't offer an alternative then I'm not using your service.