DSS 2026 — Onchain Security Markets in a World of Offensive AI John Ennis Source video: https://www.youtube.com/watch?v=Sku111HOIgw Automatic captions — not human-reviewed. Complete spoken-audio coverage is unverified. Source language tag: en (English). Source export SHA256: 8ccb8156078d08c1ea936ec3bb24b6ed7009bb5ec8e81b94b3963423d834e174 [0:01] Good. [0:02] Tired. [0:04] And before I start um on this next one, [0:08] I'll press my road to see if my slides [0:09] work. Yes, I do. [0:11] Um can I get a show of hands here? [0:13] Anyone who has a safe multisig just so I [0:15] know [0:16] kind of the rods. Okay, so we're looking [0:19] at [0:19] maybe half, maybe half. [0:21] Um [0:22] if anyone isn't aware of Safe is a [0:24] multisig um and [0:27] I work I've been working with Safe for [0:28] about maybe six years. I originally [0:30] joined Gnosis and we when I joined [0:33] Gnosis, they were actually originally [0:35] working on the [0:36] prediction markets and [0:38] I think there's still we we you're [0:40] probably familiar now with Polymarket. [0:42] Um they brought prediction markets to [0:43] the main to the mainstream, I would say. [0:46] But there's still a bit more innovation [0:47] that could be done with regard to [0:48] prediction markets, specifically in the [0:50] security space, and I want to talk to [0:51] you about kind of our insights that [0:53] we've gotten from this and kind of our [0:55] journey on where we're kind of releasing [0:56] Safe Safe Net and uh putting together [0:59] some on-chain security markets to tackle [1:01] the [1:02] the rising threat of offensive AI and [1:04] the fact that yeah, [1:06] existing threat actors are just more [1:07] scalable now with AI than before and [1:11] yeah, we need to level up as a result. [1:14] This used to be 2.26 million so far in [1:16] hacks, but by I mean by by Binance uh [1:19] increased this significantly as well. [1:22] Um this is so far we're looking at 8 [1:24] billion worth of losses. Everyone knows [1:26] what we're facing here, uh but really [1:28] we're looking at deep fakes as as a [1:31] rising trend. There is autonomous [1:34] agentic exploits happening, constantly [1:35] reviewing patterns around your codebase [1:38] and also assisted social engineering and [1:40] most of all like social engineering is [1:42] kind of the number one cause that we're [1:43] seeing that is contributing to these [1:45] numbers. [1:48] Someone had a good talk yesterday around [1:50] how you can [1:52] kind of triage your agents to look at [1:54] different areas of the codebase so they [1:55] don't default to one when they're [1:56] finding bugs. [1:58] While you're doing that and taking those [2:00] insights, can also kind of create other [2:02] agent roles itself to simply recursively [2:05] check what bugs that other agent your [2:08] other agents are finding and [2:11] chain attacks together, chain exploits [2:13] and continuously run. And if you can't [2:15] find an exploit in someone's protocol, [2:17] be it on the front end, back end, [2:18] contracts, or PC level, come back to the [2:21] bug finder agents and the agents again [2:24] recursively continue to find new bugs [2:26] and eventually [2:27] you'll get a successful exploit. [2:30] Previously this was done by just [2:31] security analysts alone, but now with AI [2:34] it's just more scalable. You can just [2:36] get hundreds of agents working on this. [2:39] So the threat actors now just have yeah [2:41] more scalable tools than previously. And [2:44] what happens as a result of that is [2:47] the common vulnerabilities and exposures [2:50] available [2:51] found in all stack, be it open source or [2:54] closed source, just continuously [2:56] increases. So we're looking at since [2:58] like AI like LLMs kind of began to reach [3:01] mainstream adoption, we just see [3:04] doubling and doubling doubling in terms [3:05] of common vulnerabilities. And [3:08] we're already 72,000 in Q3 [3:11] in. So I'd say we'll hit 100,000 this [3:14] year. And again that's doubling from [3:16] last year. So as you can see this is [3:18] just becoming [3:20] you know just [3:22] essentially exponentially [3:24] increasing. And again it's [3:26] agents have made these existing threat [3:29] actors more scalable. [3:32] So yeah we're kind of in a bad situation [3:34] at the moment. [3:35] We're basically like trying to kind of [3:37] stack up some some defenses quite [3:40] quickly. [3:42] David had a good talk on that just just [3:44] before around [3:46] yeah how how they're doing that. But [3:48] what since the since LLMs are [3:50] essentially using crowd sourced [3:52] knowledge [3:54] and attack attackers are using AI to [3:56] leverage the crowd to attack, [3:58] we can leverage the crowd also to [3:59] defend. And that's the kind of insight [4:01] that we wanted to in into our next step [4:03] product that we're building. [4:04] So how exactly do we want to apply this? [4:06] So like if they're using, you know, the [4:08] crowdsourced knowledge, how can we [4:09] defend a crowdsourced knowledge? [4:11] Um and where we want to to actually [4:14] address that is in specifically [4:16] transaction security. [4:18] Because the biggest thing causing these [4:20] hacks and these exploits is that you [4:23] don't really know what you're it's hard [4:25] to actually match what you're signing [4:27] with what you intend to do. [4:29] And that's happened across the board for [4:31] all the recent hacks that involve [4:33] billions or millions and [4:36] that is still a very it's I think it's [4:37] what I think it is one of the biggest [4:39] problems. It's a multifaceted um [4:41] security space with a cat and mouse [4:42] game, but I think this is one of the [4:43] biggest challenges and one of the [4:45] biggest problems we're all trying to [4:46] solve. [4:47] And the kind of scope today or the kind [4:49] of gaps we see is [4:51] we tell people, oh, verify what you're [4:52] trying to sign with a third-party tool [4:54] or use our company [4:56] um to do this and we'll we'll review all [4:59] your transactions and we'll tell you [5:01] whether it's good or bad. And that's [5:03] fine, but there's gaps in that as well. [5:06] And the gaps that we see is that [5:09] one, advisory warnings are structurally [5:11] useless. So if you if you give someone a [5:13] warning to say what you're signing is it [5:17] might be malicious, [5:18] people can just click sign and do it [5:20] anyway. Structurally they can just [5:21] bypass it or if your UI is compromised, [5:25] that that warning's not coming through. [5:27] You are going to essentially you you [5:30] have no last line of defense. You [5:31] misinterpret the UI, you sign, and [5:34] you're you're wrecked. [5:35] Um [5:36] the second as well is that [5:38] again, self-custody is irrevocable. I [5:40] think we all know this as well, but [5:42] again, the attacks are getting more and [5:43] more sophisticated. So it's getting even [5:45] more like the cost like the cost of [5:48] attack is getting lower and yeah, [5:50] obviously the honeypots are getting [5:51] bigger. [5:52] And [5:54] we've we've noticed this as well. [5:55] There's no real economic accountability [5:58] really for these third-party tools or [6:01] third-party companies that are um [6:03] checking transactions. Like ultimately [6:05] it's the that's using the wallet is [6:07] going is the one that gets wrecked if [6:10] someone didn't catch [6:12] a threat. So if you have a company [6:14] that's checking your transactions and [6:16] says, oh, we didn't catch that, but [6:18] you're the one that's facing the [6:19] millions worth of hack millions worth of [6:22] dollars worth of hacks, it's that's [6:24] challenging, right? Like that's kind of [6:26] in a way that's kind of unfair, but [6:28] that's kind of how it exists right now. [6:30] Of course, there's legal recourse, but [6:31] really there's not as much skin in the [6:33] game, and we found that ourselves as [6:35] well. [6:38] So what we have built now with what [6:40] we're what we've released now actually [6:42] with Safe Ness is what we want to do is [6:45] multi-party security checks. So we want [6:47] to leverage the crowd and not just [6:48] select one single actor that is kind of [6:51] doing the security checks for our [6:53] wallets. [6:54] We want to leverage multiple parties. We [6:56] want them to be accountable. We want [6:57] them to put up bonds as a result of each [7:00] transaction that they want to [7:01] participate in. [7:02] And we want to enforce it on chain, [7:04] which means it's not just a UI warning [7:06] that you can just bypass, that it needs [7:08] to be a transaction needs to be attested [7:10] on chain in order for you to execute it [7:12] as well. [7:13] And we want to address specific risks [7:16] that these multiple parties we call them [7:18] sentinels. We have four sentinels right [7:20] now. And we want the sentinels to be [7:22] addressing specific risks. [7:24] Address poisoning, unauthorized approval [7:27] to your contracts or to your tokens, [7:29] constantly checking malicious contracts [7:31] and whether they are compromised at the [7:33] time of the transaction, [7:35] hidden ownership changes, so changing [7:36] the actual ownership underneath without [7:38] you realizing. This can be used through [7:40] like unsupported delegate calls or or [7:42] nasty like hidden delegate calls. [7:45] And just generally trying to manipulate [7:48] again what you're trying to do with what [7:50] they want you to do as well. [7:53] And we have this charter, which is [7:55] essentially a dynamic rule book, let's [7:57] say a court of law. Since there's always [7:59] new laws coming through or there's [8:00] always new attack patterns, so we'll [8:02] always be updating this kind of charter, [8:03] this court of law of what to you need to [8:06] address or the sentinels need to [8:08] address. [8:10] So just to walk you through how this [8:12] would work, [8:13] um [8:14] we have [8:16] you propose a transaction and we'll walk [8:18] through the happy case. We propose a [8:19] transaction. There's sentinels. Right [8:21] now we have four, but really we want [8:23] more sentinels and we actively want [8:25] anybody that's operating in this space [8:28] to potentially evaluate running a [8:29] sentinel. [8:30] Um the sentinels check against the [8:32] charter. Are there specific risks or [8:35] specific rules that this transaction's [8:36] breaking? [8:37] If if it's unanimously secure, they all [8:40] say thumbs up, it's good. The validators [8:42] are more of an admin role. They'll [8:44] attest the transaction on chain. They do [8:46] this thing called a frost signature, [8:48] which is just again they all participate [8:50] in kind of generate a cryptographic [8:51] signature. And once that is attested, [8:54] it's broadcasted cross cross chain and [8:56] then the guard can verify that yes, this [8:59] user can actually execute this [9:00] transaction because the validators have [9:02] created the frost attestation and the [9:04] sentinels say this is secure, we back [9:07] this transaction. [9:09] Um that's if you enable the guard. You [9:11] don't have to enable the guard. You can [9:12] just simply use warnings instead, but if [9:15] you do enable the guard, you have to go [9:16] through a delay and many people in the [9:18] previous talks talked about having [9:20] delays associate with critical [9:21] operations or delays associate with [9:23] critical like large-scale transactions [9:25] that that's true and a lot of hacks [9:27] could have been prevented or not [9:29] prevented, let's say, but it could have [9:31] added another layer of defense if there [9:33] was delays incorporated as well. [9:36] For the insecure case, if one sentinel [9:38] says this is insecure and this is the [9:40] reason based on the charter why this is [9:42] insecure, that will be reported to the [9:44] user. You will The validators will not [9:46] attest and you have to go through delay [9:49] if you have the guard enabled. [9:53] Right now the validators who are [9:54] performing the administrative role, this [9:56] we have like Nosis Safe, we have [9:58] Greenfield, we have core contributors [10:00] running validators, they're generating [10:02] frost signatures, [10:03] and the sentinels they're putting up [10:05] bonds per each transaction [10:07] um that they want to participate in and [10:09] we have the of Blocksec, OpenCover, [10:11] Candide, and TenGard. Again, we want [10:13] more validators and we want more [10:15] sentinels to make this truly kind of [10:17] permissionless system [10:19] and build up because it's a permission [10:20] system right now, but this is what we're [10:21] working with right now. And yeah, happy [10:24] to chat to anyone after if you're [10:25] interested in looking into this deeper. [10:30] Again, this is live right now in Safe [10:32] Wallet. You have to subscribe to Safe [10:33] Pro, but [10:35] users simply, you know, [10:39] either pay pay a subscription or they [10:40] pay on a per transaction basis. [10:43] The sentinels get that transaction, they [10:45] get the fee as well. If the sentinels [10:46] participate, they get they get a portion [10:48] of the fee. [10:49] And again, they're they have bonded [10:52] verdicts. I don't have enough time, but [10:53] if the sentinels disagree and one of the [10:55] sentinels is wrong after the fact, they [10:57] lose their bond and that bond is [10:59] distributed to the other sentinels. So [11:01] the game theory works that the sentinels [11:03] want to find things that the other [11:06] sentinels disagree with, [11:09] which is actually hacks like because [11:11] like malicious transactions tend to be [11:13] the 0.00% [11:16] of transactions. So you want the game [11:18] theory to work that sentinels find [11:20] specific things and get the bonds of the [11:23] other sentinels as well and of course [11:26] participate so they're getting the fees [11:28] on a on a constant basis as well. [11:31] Again, the validators do they perform [11:33] the frost signature, they create the [11:35] attestation in the guard, and then yes, [11:37] the user [11:38] can execute a transaction if they have [11:41] the guard enabled. [11:43] And here you can just see right now this [11:45] here just creating a safe transaction. [11:47] Sorry, I don't know why I'm looking at [11:47] this. I know these slides. [11:50] You can see the you can see someone [11:52] create a safe transaction. The sentinels [11:54] found no issue [11:56] with that transaction, so you can simply [11:59] go ahead. That's the happy case execute. [12:01] If the sentinels did find something, [12:03] this would be reported. You would have [12:05] one sentinel reporting something or [12:06] many, and you can click on the explorer [12:09] and here the explorer will tell you [12:12] who said, who said this is actually [12:15] insecure or not secure, [12:17] and what's the rationale of why they [12:19] said insecure. If So here you can see [12:23] block six said secure. Candid and 10 [12:25] cards denied it and said insecure. [12:27] And here this is on all on chain, so it [12:30] has to be quite a rigorous explanation. [12:33] But we will summarize this in the UI, [12:35] but ultimately it fails because you have [12:39] a wrapped ETH. [12:41] You've created an unlimited approval to [12:43] the wrapped ETH contract with that [12:44] account. We don't want unlimited [12:46] approvals in this day and age with [12:49] with specific like, you know, basic user [12:51] operations. So yeah, this is labeled [12:53] insecure as per the charter. [12:57] So why why why do we think this is [12:59] different from the existing kind of [13:00] state of play right now? It's the fact [13:02] that we have multiple parties competing. [13:04] Like right now in like the state like we [13:07] have multiple attack [13:10] like [13:12] attack participants or let's say [13:13] malicious actors all competing. [13:16] Whereas it doesn't really have that [13:17] same, you know, game theory for the [13:20] people who are trying to defend. It's [13:22] all kind of like you're on your own [13:25] mentality. So we believe that like that [13:27] the we want to leverage the crowd here [13:29] and leverage the experts and the best is [13:31] to actually have multi-party competition [13:33] as a result of this. [13:35] Again, no just warnings that you can [13:37] bypass on chain enforcement. So if the [13:39] UI is hacked, you can still actually, [13:42] you know, [13:43] have critical transactions blocked on an [13:46] on chain level. [13:47] We have a dynamic charter. Like the law, [13:49] there's always new laws or new attacks [13:51] arising. We'll also have We have a [13:52] charter as well that's always being [13:54] updated with new attacks that are being [13:56] created because this is a cat and mouse [13:57] game at the end of the day. There's no [13:59] one solution. Not This This won't fix [14:01] everything, but it will be another layer [14:03] of defense. [14:05] And again, making making people [14:06] accountable actually, making people [14:08] accountable for self for not finding [14:10] things or finding things and reporting [14:12] them wrong. [14:14] And yeah, we believe this is in the [14:16] direction of creating more [14:17] accountability for uh yeah, defending [14:20] against these threats. [14:23] In the future, we want to create a more [14:24] open [14:25] sentinel market, open validator market. [14:28] Um, and again, we want to make them [14:31] specific with transaction security. We [14:33] want to support multiple accounts, so [14:34] this is not just a safe solution. I'd be [14:36] happy to talk to any wallet providers or [14:38] any security companies that would love [14:40] for their for our charter to be updated [14:42] to their um [14:45] uh structure. And yeah, we can also [14:47] create insurance products on top if the [14:49] sentinels can guarantee that they will [14:51] check these things as well. [14:54] That's me done. Um, I would just say you [14:57] can scan this QR to learn more. If you [14:59] want to reach out to us directly about [15:02] running a wallet, running a sentinel, [15:04] running a validator, you can get this QR [15:06] code. And also, we're in Mumbai Cricket [15:08] Club uh on [15:10] Wednesday. [15:11] Um, and yeah, we have a we have an event [15:13] there with the Ethereum Foundation um if [15:15] you'd like to attend. So scan that QR [15:17] code if you'd like to attend the event [15:18] as well. [15:19] And yeah, I think I'm all out of time. [15:22] So um yeah, [15:23] I don't chat to me if you have any [15:25] questions after. I can't I don't think [15:27] I'll I don't want to run into anyone [15:28] else's time. [15:30] Thank you.