DSS 2026 — Forging PLONK Proofs Against an Unchecked Verifying-Key Cache (gnark, BN254) Vijaykumar Singh Source video: https://www.youtube.com/watch?v=y-XAdd1iYUQ Automatic captions — not human-reviewed. Complete spoken-audio coverage is unverified. Source language tag: en (English). Source export SHA256: 98ca1532f8a16b6d89cf4e2ed2b7e79e402a94cc867e48367b7d6a2555f6b8a8 [0:00] Shall we? Okay. So, [0:03] I'm going to talk about forging Planck's [0:05] proof against an unchecked verifying [0:07] key. [0:08] This is the setup in [0:12] So, [0:13] this work was done by my coworker at [0:16] Open Zeppelin, [0:17] Vaslin, [0:19] aka Vesco. [0:22] And yeah, I'm here to present it. All [0:24] right? Okay. [0:25] So, what is this [0:27] this setup about? [0:29] Before I go further, these are the [0:31] things I'm going to cover very quickly. [0:33] So, I'm going to describe the problem [0:35] for you. [0:36] Then I'm going to go through the three [0:37] findings in the [0:39] one is [0:40] zero cash, crafted cash, both are [0:43] both give you a proof forgery, whereas [0:46] the the last one is a degenerate base [0:48] point, which is a statement forgery. All [0:50] right? And then we'll talk about [0:51] disclosure and and conclusions. [0:55] Okay. So, [0:56] I'm going to spend a bunch of time on [0:58] 2.1 and you know, rest we see if we get [1:01] time. [1:03] All right. So, what is happening here? [1:07] So, let's try to build some context. [1:10] So, in BN 254 curve, you basically have [1:13] KGZ commitment schemes, right? [1:16] And [1:17] it's a pairing-based [1:19] commitment scheme where [1:22] the end verification is essentially [1:25] [snorts] checking some pairing [1:27] equation turns out to be one or not. [1:30] So, essentially, if you think about it, [1:32] you can see that pairing check [1:35] corresponds to [1:37] you have [1:39] you have E of A one. [1:42] So, let me see what what one is. One is [1:44] the image. So, [1:46] if you recall what pairing is, pairing [1:48] is a map from [1:50] G1 cross G2 to, you know, FP12 in case [1:53] of eight pairings, which is what BN 254 [1:55] curve uses. [1:57] And [1:59] so one is generator for G2 group [2:03] and [2:04] your X is the the toxic waste which is [2:07] produced when you you know you do SRS uh [2:10] so the initial setup ceremony. [2:12] So in the end you you you check [2:15] essentially whether this pairing map [2:17] holds [2:18] uh sorry this pairing equation holds uh [2:20] and [2:22] if it holds then then you pass the proof [2:24] uh if it doesn't then it you know the [2:26] full proof fails. Okay so this is a [2:28] setup. [2:29] Now uh pairings are bilinear maps so [2:32] essentially you can uh what you can do [2:34] is you can move this X here you get XB [2:36] you have A so AXB [2:39] uh [2:39] should turn out to be so A + XB should [2:42] turn out to be identity. Okay [2:44] so if A + XB is identity then then [2:46] essentially you have the the the the map [2:49] will check out all right. [2:51] So it all rests on one element really of [2:54] G2 uh which is the X uh the toxic waste. [2:57] It is unknown to prover it's unknown to [2:59] verifier yeah but they they have both of [3:02] them have the commitment to X. [3:04] Okay [3:05] so [3:06] so ideally um [3:08] so this is a setup [3:10] but [3:12] unfortunately [3:14] quote unquote unfortunately the grow [3:16] verifier [3:17] instead of basically checking against X [3:20] checks against something else which is [3:22] the line [3:23] in the in the [3:25] so let me just clarify what the line is. [3:29] Uh so I'll come to the line a moment but [3:31] first of all let's see uh what happens [3:33] um [3:34] so if you recall from the previous slide [3:37] we had this uh [3:39] equals to identity [3:41] right if you go on the next slide [3:43] so pairings are are are are are [3:45] non-degenerate but if you have EA comma [3:49] zero element [3:51] of the G2 yeah zero is the identity if [3:54] you're thinking about elliptical group [3:55] as as an additive group right so if you [3:58] do E of A zero then it will always come [4:00] out to be one [4:02] No matter what. [4:03] So, if you look at this map here, [4:07] if I had zero and zero here, [4:10] it will turn out to be [4:12] it will pass, right? Irrespective of A [4:14] and B. Doesn't matter what A and B are. [4:16] So, can we have that setup? [4:18] All right? So, can we have Q zero and Q [4:20] one as basically [4:23] the additive identity or or the [4:26] so-called point at infinity for the [4:27] elliptic curve. That is a problem [4:30] So now, so the question is, can we have [4:33] that? [4:36] And the answer is yes. We can have that [4:40] because of the following scenario. So [4:41] what happens is if you look at the [4:43] pairings, [4:44] the way pairings are computed for the [4:46] GNARK is that you have E of P comma Q. [4:49] The Q is fixed. Yeah, that was from the [4:51] toxic waste in in the second part. [4:54] So what what what what this [4:58] computation of pairing does is that [5:01] the it runs a Miller loop. [5:03] All right? [5:04] So and in the Miller loop, what happens [5:06] is [5:08] there are already some precomputed lines [5:10] which are inside the Miller loop. [5:12] Okay? So instead of having using [5:15] commitment to X, [5:17] yeah, the lines which are already [5:19] precomputed are part of your of are part [5:22] of your computation. Okay? And this is [5:24] where the problem comes in. We'll see [5:25] that in a moment. [5:26] So Miller loop runs and then, you know, [5:28] there's a final exponentiation and then [5:29] you get if you get one, you pass. If you [5:32] get do not get one, it fails. [5:34] So the [5:35] so the idea is [5:39] the following. So [5:42] what [5:45] GNARK library does is it says that it's [5:48] checking for this pairing [5:50] where you have basically A of G two, [5:53] which is your toxic waste. But [5:55] essentially what it does is [5:57] it has a verifier already precomputed. [6:00] The lines [6:01] precomputed given by the prover. [6:04] It does not checks it. [6:06] And then [6:07] if these lines are not properly [6:09] constructed, [6:10] then there's the issue. Okay? So let's [6:12] look at this [6:14] So the idea is that we have [6:17] lines, we have precomputed lines and [6:20] they are never enforced when they are [6:22] used in the proof. All right? So so the [6:25] the starting the [6:27] the issue is at the [6:29] even before the start of the proof. It's [6:31] in the setup. [6:33] The the lines are precomputed. If the [6:35] verifier does not checks it [6:38] against it being identity all the time, [6:40] you basically have the have the fake [6:41] proof passing. So so this is the [6:44] first issue. [6:46] All right. So that's that. [6:50] Okay. So second proof is the following. [6:52] Second issue is the following. [6:54] Let's go back to our equation again. [6:57] We have here [6:59] AX so we had A + XB equals to identity. [7:03] So earlier the issue was what? The issue [7:05] was basically you take [7:07] you took X to be commitment X commitment [7:10] of X to be identity and [7:12] the commitment of one to be identity. [7:15] Again, do not confuse it. By here [7:17] identity I mean the zero element and [7:18] that was a problem. [7:20] Now the second case can arise if you [7:23] have X as [7:26] what you can get still [7:28] zero if you had X as [7:30] some kind of [7:32] minus A + B. [7:34] Oh sorry, minus A divide by B. Okay. So [7:36] that's the [7:37] the source of the second issue [7:40] as you can see here. [7:43] So the idea is you have a toxic waste. [7:47] The commitment here. [7:48] So the issue is that the attacker will [7:51] what will do is instead of [7:53] it will it will take some X prime of its [7:55] choice in a way that [7:58] eventually you you have you turn out to [8:01] be this you have turn out to have this [8:03] equation A + X prime B [8:06] and then you choose B to be minus A or X [8:08] prime and then you you essentially have [8:10] what? [8:11] Again a proof forgery, right? Okay. [8:14] So in both the cases [8:17] what was the problem? [8:21] So in first case you you you don't want [8:24] the zero of the elliptic curve or the [8:27] point at the infinity [8:28] to be the part of the proof. So you [8:29] should check that, right? So [8:38] so I'll skip the third one uh but you [8:40] can have a look uh [8:42] so let's see. So how was the issue [8:44] basically uh [8:46] so on 7th of [8:48] July the issue was [8:50] drafted and then disclosed to the um [8:54] disclosed to the uh Nark. [8:56] Uh it was confirmed and basically you [8:59] know [9:00] um [9:01] and then it was acknowledged but it was [9:03] considered to be out of scope of the [9:05] trusted model which actually makes sense [9:06] because it is before the the prover [9:09] verifier game assumes that the the [9:11] trusted model is honest on the top [9:14] but uh [9:16] and then on 28th of August it was [9:19] approved for the [9:20] public discussion. [9:22] Okay, so now why do we call it shadow [9:24] boxing? [9:26] Well, the idea is that in verifier [9:29] you are basically [9:31] seeing G two, [9:33] right? You're you're looking at X [9:35] at the commitment of X [9:36] but essentially what hits you is a line, [9:39] right? Because you have basically these [9:41] computation of lines which are which [9:43] could be malicious, right? So so so we [9:46] do the validation but against the wrong [9:48] object. Yeah. Uh yeah, so that's why the [9:51] name is called you know shadow boxing [9:53] because you are you're fighting the [9:54] wrong opponent. Yeah, the the opponent [9:56] actually is a line, right? So this is [9:58] the the whole setup of the uh [10:01] of the [10:03] of the shadow boxing um and [10:07] yep, I I think you know what? [10:09] I pretty good. I finished earlier. [10:12] Thanks, thanks for your attention. [10:15] >> [applause]