Re: Collision attack against long Key Ids
Andrew Gallagher via Gnupg-users <[email protected]> Wed, 11 Mar 2026 18:36:02 +0000
| Newsgroups | gmane.comp.encryption.gpg.user |
|---|---|
| Message-ID | <[email protected]> |
On 11/03/2026 16:57, Nombre y Apellidos via Gnupg-users wrote: > > I see a blog post about collisions in key Id's in this blog: > https://soatok.blog/2026/01/07/practical-collision-attack-against-long-key-ids-in-pgp/ As usual, Soatok wraps technical correctness in hyperbole and self-promotion. > According to the article itself, it can lead to cases of usurpation of > signed data, such as software packages. > > Is it really a weakness in PGP/GNUPG? It's a weakness *if* people assume that key IDs are unique identifiers, rather than a convenience. In older user guides, key IDs were treated as unique identifiers but this has not been recommended for a long time. You will note that Soatok references a decade-old stackexchange answer, and even that text recommends the use of fingerprints instead. In the openpgp protocol itself, the only place that key IDs are used is as a label to help a receiving implementation find the correct key for decryption (or in older messages, signature verification). The security of openpgp has never relied upon key IDs though. To exploit a birthday attack against key IDs, an attacker would have to create two colliding keys, then use one for innocent purposes, get other people to trust it, then trick them into using the second one instead, and hope they only check the key ID and not the full fingerprint. But remember that the attacker *already controls both keys*. Swapping one key under the attacker's control for a second key under the same attacker's control doesn't get the attacker any privileges they didn't already have after convincing people to trust the first key. The best that Soatok can do is suggest that the colliding keys would give an attacker plausible deniablity after the fact, in the absence of other evidence. But given that a birthday attack is known to be feasible and a preimage attack is known to be technically infeasible (for now), nobody with any sense will believe them. Plausible deniability is as bulletproof as a piece of wet tissue paper. tl;dr: don't panic. A