DSS 2026 — The State of DeFi Security - 2026 Peter Kacherginsky Source video: https://www.youtube.com/watch?v=_l0JuHU2roQ Automatic captions — not human-reviewed. Complete spoken-audio coverage is unverified. Source language tag: en (English). Source export SHA256: b8aa4734d8650bed1bc690314b5f31f5527a59cb96b4911ec95145aa9d512661 [0:00] All right. [0:01] Welcome, everyone. Namaste, India. It's [0:04] so good to have every single one of you. [0:07] Welcome to the state of DeFi security. [0:09] It's an honor to give the talk now [0:11] fourth year in a row where we look back [0:13] at the industry and learn. [0:18] So in case we haven't met, my name is [0:19] Peter Kacherginsky. I'm the editor of [0:21] the Blockchain Threat Intelligence [0:23] newsletter for the past eight years, [0:25] and also a security advisor to some of [0:27] the projects out there. [0:30] So before we begin, a quick [0:31] announcement. In light of everything [0:33] that was happening in the past two days, [0:36] I decided to put the Blockchain Threat [0:38] Intelligence newsletter out out there as [0:41] a public good, freely accessible, so you [0:43] can subscribe, learn, and hopefully get [0:46] the lessons to not get hacked. [0:51] So let's begin our story. Usually when [0:53] we gather here, we jump into the [0:55] statistics and the threat models and all [0:58] the case studies, but I want to do [1:00] something different today. I want to [1:02] tell you a story. [1:04] It's a Western. [1:06] It has a nasty two gangs operating [1:09] there. There are lots of robbed banks, [1:12] and some of you are in it as well. [1:22] Maybe the slides. [1:25] There we go. [1:27] So to get to get an idea of how we got [1:29] to these pretty tough times, we have to [1:31] start at the beginning, the drought. [1:34] Our story begins in the winter of [1:36] 2025-2026, [1:38] just as we all coming down from the [1:40] hangover of the bull market of the past [1:42] year. [1:43] Bitcoin just hit 60k, [1:45] and all of the projects out there are [1:47] starting to really tighten their belts. [1:49] The budgets are getting cut, especially [1:51] security budgets. Audit jobs and [1:54] contests and other security work that [1:56] used to be a plenty in the past year [1:58] really going down. [2:00] Suddenly all of the security researchers [2:02] out there who used to be really busy and [2:04] could make a decent living find [2:06] themselves unemployed without a job on a [2:08] curb with no future prospects. [2:13] Some of the smarter of you switched over [2:15] to bounty hunting and that worked for a [2:17] little while until [2:18] bounty programs, just like the rest of [2:20] the world, were overwhelmed by the AI [2:23] slop and started introducing [2:25] various types of friction, especially [2:28] the submission fees. [2:31] And just to give you a context of the [2:33] world, both inside crypto and outside [2:35] crypto, what was happening with the AI [2:38] craze invading everyone, [2:40] managers and founders realized their [2:43] ultimate goal of having employees who [2:46] don't sleep, don't eat, don't require [2:49] any asks and just work, work, work all [2:52] day. [2:53] So AI is going to write the code, test [2:56] the code, audit the code [2:59] and that's how the layoffs begin all [3:03] around the world. [3:06] And just around that same time when the [3:09] whole AI seems to be taking over the [3:11] world, the frontier models introduce [3:13] excellent models like Opus 46 and GPT [3:17] 53. [3:18] Where a year ago these models could not [3:19] even tell a difference between [3:21] re-entrancy and price oracles, suddenly [3:24] you could get ready-made foundry POCs [3:26] that you could just run on the main net [3:28] and hack whatever you want to. [3:32] And it is in this context where you have [3:34] a whole bunch of security researchers [3:36] who are well skilled [3:38] and now very well equipped out of the [3:41] job [3:42] in a down market, [3:43] what do they do? [3:46] Well, a lot of them they left the market [3:48] altogether, they moved on to other jobs, [3:50] but some of them they burned the whole [3:53] town down. [3:54] We went from three or four hacks a week [3:57] to three or four a day. [4:00] The last time we met here in in [4:02] Argentina, we had just 24 hacks a month. [4:06] By April, as we called bloody April, [4:09] we hit 67. [4:11] That's more than double. [4:14] And how is everyone getting hacked? [4:16] Well, guess what? AI is really good at [4:19] finding bugs like signature [4:21] verification, which wasn't even [4:23] registering in the past years. [4:25] Seeming inability by projects to [4:28] properly verify the message integrity. [4:30] They don't know how to check for [4:31] nonsense, what should be in it, what [4:33] shouldn't be, what can be replaced. And [4:35] then we have devastating consensus [4:37] failure bugs. Those are the type of bugs [4:39] where it attackers figure out a way to [4:41] print infinite amount of money from [4:43] bridges and chains. And we had that [4:45] particular bug class in the past, maybe [4:47] one a year. This year we had 10. [4:51] And even more were responsibly disclosed [4:53] and patched before they were getting [4:55] exploited. [4:56] Such as the Zcash earlier this year and [4:59] the Ripple infinite minting bug just a [5:01] month ago. [5:02] And of course, the dreaded price oracle [5:04] manipulation. We've worked for years to [5:07] bring down the risk of this very well [5:09] understood and known attack class, but [5:11] driven by all these tiny little old [5:13] hacks from projects on mostly Binance [5:15] Smart Chain, [5:16] this shot right back up to the top. [5:20] And who is getting hacked? So again, [5:22] looking back in November of last year, [5:24] those are mostly projects in the sweet [5:26] spot between 100k and a million dollars, [5:28] making it worthwhile for attackers to [5:30] spend time understanding the protocol, [5:32] weaponizing their bugs, and so on. [5:35] By April and September and it continues, [5:39] most of the hacks were under 100k. [5:42] Writing exploits, looking for bugs, [5:44] weaponizing them, [5:46] and executing them became so cheap that [5:49] whatever little project out there had [5:51] with some TVL is now a worthwhile target [5:54] for attackers. [5:58] What was unusual [5:59] back in April is that it's been a few [6:02] months since we heard from our North [6:05] friends. [6:06] Were they taking a vacation? [6:08] Nah. They were just busy casing their [6:11] next targets. [6:12] Their first target was the Drift [6:13] protocol, which unprecedented was [6:16] targeted first in person at a conference [6:19] just like this one, where they were [6:20] approached by someone to build trust and [6:23] then later on build a relationship to [6:24] send malware. So be very careful who you [6:27] interact with at DSS and DEFCON in the [6:29] coming days. [6:30] And then of course, we have a classic [6:32] trader-trader tactic of compromising a [6:35] an engineer laptop at layer zero to [6:37] eventually deploy a pretty sophisticated [6:39] exploit against layer zero [6:44] server nodes. [6:47] We'll come back to these guys more. [6:49] But what I want to bring up is that [6:50] around April time [6:53] Frontier Models, Frontier Labs realized [6:55] the security sells. [6:57] And how are they going to monetize it? [7:00] Well, through scarcity and guard rails. [7:03] To protect us all from ourselves, the [7:05] latest models like Mythos were now a [7:08] gated product where you have to be on a [7:10] special list and you have to know [7:11] somebody in Anthropic and then you have [7:13] to sign up for an extra expensive API [7:15] plan to get access to a highly [7:17] token-hungry [7:18] model. [7:20] For the rest of us, we were forced into [7:22] this guarded garden to make sure that [7:25] God forbid we do something security [7:27] related. [7:29] What was interesting on top of that as [7:30] we learned from the Hugging Face [7:32] incident is that those guard rails not [7:35] only apply to [7:37] potentially offensive [7:40] security work, but also a completely [7:42] legitimate work. When Hugging Face [7:44] engineers tried to use the very same [7:46] model that was attacking them to defend [7:48] themselves, they were told, nah, you're [7:50] doing something suspicious. [7:52] Lesson learned. [7:57] By July of this year, both the guard [7:59] rails and also just going through all of [8:01] the low-hanging fruit contracts that [8:03] were out there was having some effect on [8:05] the number of hacks. So we briefly [8:07] dipped to around 51 hacks. [8:12] But then what happened? The Chinese [8:14] models arrived. Models like GLM 52 and [8:18] DeepSeek V4 and all the Quens 36s and [8:21] 38s and Next and so on [8:23] suddenly had very similar, not the same, [8:27] but good enough cyber capabilities to [8:30] bring all the attackers back into the [8:32] game. [8:35] Most importantly, [8:37] the models did not have guard rails. And [8:39] whatever guard rails were put in there [8:41] were quickly removed through [8:43] obliteration and uncensored [8:45] versions of those same models. [8:47] Even more interesting, a lot of the [8:48] security researchers started publishing [8:50] specially post-train models like [8:52] Cantina's Apex that are specifically [8:54] designed for doing security work. [8:58] In the end result, [9:01] by September of this year, we not only [9:03] matched the numbers of Bloody April, we [9:05] surpassed it. [9:07] If you want to connect the dotted lines, [9:09] unless something changes in the near [9:11] future of how we do security, [9:13] we should see around 95 plus hacks by [9:17] March of next year. [9:21] What's even more interesting [9:23] are the type of hacks that are started [9:25] happening. [9:26] We started with the low hanging fruits, [9:28] all those price oracles and just [9:30] contracts don't know how to implement [9:31] burn functions over and over again. [9:33] There's the same bot that I keep [9:34] watching. There's just [9:36] pillaging the whole Binance chain. [9:39] Once those were done, attackers started [9:41] moving down the stack to more and more [9:43] complicated targets. They started with [9:45] bridges. They continued to whole chains. [9:49] We had the Liquid Network hack, $320 [9:51] million. [9:52] They later moved on to wallets where it [9:55] seems [9:56] every day we hear about a weak PRNG or [9:58] some other cryptographic issue of how [10:00] wallets are getting generating their [10:01] seed. [10:03] And they continue going down this phrase [10:04] to down this the stack until they're [10:07] hitting signing infrastructure and so [10:09] on. [10:12] So I think folks, as we gather here [10:15] today, it's time for a reckoning of all [10:18] the things that broke [10:20] this year. [10:23] On the one hand, [10:24] we have the $8.7 billion number [10:29] that was stolen so far this year. [10:32] And the vast majority of that is North [10:35] Koreans just pillaging our ecosystem, [10:39] exploiting weak operational security, [10:43] security programs that cannot withstand [10:45] a single laptop compromise. [10:49] On the other hand, we have 574 hacks [10:52] this year so far. That's almost double [10:54] to the same time period last year. [10:57] And that's I call them the the swarm of [11:00] AI hackers who are just exploiting [11:03] anything that was not audited, did not [11:05] have the right code coverage, [11:07] and more interestingly, did not have the [11:09] right incentives for security [11:10] researchers to do the right thing before [11:12] them. [11:14] And when I say incentives, [11:16] this is what bug hunting looks like [11:18] today. [11:20] A good number of you switched over to [11:23] the dark side [11:25] where it is now more preferred action to [11:28] negotiate the bounty after the hack, [11:31] where you name your own [11:34] bounty percent, like 10%, 20%, whatever, [11:37] arbitrary, [11:38] instead of doing the right thing and [11:39] falling the programs, paying the [11:41] submission fees, maybe waiting a year, [11:44] negotiating back and forth, and then [11:46] eventually maybe getting something or [11:48] more likely just downgraded to [11:49] informational. [11:53] And how do we get there? [11:55] Well, between bounty programs shutting [11:57] down, uh platforms rather shutting down [11:59] altogether like Code Arena, rest in [12:01] peace, [12:02] to [12:03] remaining bounty programs simply [12:06] becoming hostile to ethical security [12:08] researchers. [12:10] If you are a researcher today, you're [12:12] doing this mental gymnastics where [12:15] you're thinking, well, [12:17] I could submit this critical that I [12:18] discovered, [12:20] but do I want to risk the submission fee [12:22] and do I want to invest the next year [12:24] dealing with all this back and forth, [12:27] becoming a lawyer more so than a [12:28] security professional? [12:31] And I think a lot of them are just [12:32] choosing not to submit. [12:34] And I get it. [12:36] A year ago, [12:38] you you were overwhelmed [12:40] by the AI slop. There were too many [12:42] submissions, and I personally know so [12:44] many of you triagers out there who just [12:46] burnt out. [12:47] But that was a year ago. [12:49] You had a year [12:51] to learn and use the same tools that are [12:53] overwhelming you to improve the triage [12:56] process because that is your only job to [12:58] broker between ethical security [13:00] researchers and vulnerable projects that [13:03] also need that intelligence, and you're [13:05] failing at this mission. [13:09] We're experiencing a brain drain. [13:12] The rooms I'm seeing are smaller and [13:14] smaller. I'm seeing senior and junior [13:17] talent leaving the industry for good. [13:20] On the senior side, we heard Sam Zizane [13:22] announcing that he's stepping back from [13:23] Seal [13:25] like Open Zeppelin getting acquired by [13:27] S&P, the S&P 500 people. The trail of [13:31] bids, the rumors they're negotiating [13:32] with OpenAI, and I personally know so [13:35] many of you [13:36] who left the industry altogether to join [13:39] the Anthropic and OpenAI's of this of of [13:41] the other industries. [13:44] And then on the junior side, [13:47] all those little contests, all those [13:48] little audit jobs and and bounties, that [13:53] was our school. That was the pipeline [13:55] that would bring the new junior talent [13:58] into the industry to become the next [14:00] senior security researchers. [14:03] And that pipeline is now drained. [14:07] As the market will turn around as it [14:09] always does, who will audit all those [14:12] contracts? [14:13] AI. AI is going to write the test the [14:17] code, audit the code. Do I need to spell [14:19] it out exactly how this will end? [14:23] Just ask Balancer that [14:26] had the last day and closed the closed [14:28] their doors for for good today. [14:30] Ask dozens of other DeFi projects that [14:33] already shut down this year. [14:38] And then on top of it all, we have North [14:40] Korea to deal with, who continue [14:42] operating unimpeded. [14:45] The government is not going to help you. [14:47] They are a nuclear state. Law [14:49] enforcement is not going to help you. [14:53] They don't have jurisdiction and every [14:55] sanction imaginable is already in there. [14:59] In the meantime, they're setting new [15:01] record year after year after year. [15:04] This year alone, we just had that I had [15:06] to update this slide two days ago. They [15:08] just set a new record with $6.3 billion [15:10] of a Binance hack. [15:13] And they will continue doing so unless [15:15] something changes. [15:19] Worse, we as an industry are not doing [15:22] nearly enough to stop them. [15:25] On the one hand, we have something like [15:27] Arbitrum Console who drew the line in [15:29] the sand and said, we are decentralized, [15:32] but that does not mean that you can [15:34] abuse our chain and you can use it to [15:36] transfer stolen funds. [15:39] And on the other hand, we have projects [15:40] like Thorchain who are like monks [15:43] repeating this mantra of credible [15:45] neutrality right up to the point they [15:47] themselves get hacked and then they will [15:49] shut down the chain. [15:51] Or we have someone like Circle who are [15:53] hiding behind court orders which take [15:55] weeks to obtain when transactions take [15:57] seconds or minutes to go through through [16:00] their [16:01] infrastructure. [16:07] Defenders on the other hand, [16:09] defense is so difficult. It's a [16:11] political game even more so than [16:13] technical, and they're still stuck [16:15] operating at human speed just as we [16:18] already transitioned into the machine [16:20] world. They still don't have access to [16:22] the latest models, and whatever models [16:23] they have, they have guard rails all [16:25] around them. [16:28] The last but not least, I hope you all [16:30] realize that the crypto cryptocurrency [16:33] security is kind of a canary in a coal [16:35] mine. [16:37] What we see happening in our industry [16:40] usually predates by six months or years [16:43] what happens in traditional security. [16:45] So when we had our bloody April, it was [16:48] only the matter of time until the Black [16:49] October was going to come, where [16:51] everything is getting hacked, from DNS [16:53] servers to certificate authorities, you [16:55] name it. [16:57] And before we pat ourselves on the back [16:58] saying that we're so far ahead and we [17:00] had a time to prepare for it, realize [17:03] that all of those decentralized [17:05] projects, they actually sit on very much [17:07] centralized infrastructure. And now all [17:10] of that Black October mess is coming [17:12] right back to us. Did you notice the [17:14] sudden increase in front-end [17:16] compromises? Yeah. [17:19] That's what's happening. [17:20] So if you are a chain operator or [17:22] bridge or or CeFi or DeFi and you have [17:25] some kind of back-end infrastructure, [17:27] pay attention to that, please. [17:32] So just to give you an idea of what [17:33] trajectory we are in right now, [17:35] it doesn't look so good. [17:37] Uh I had the first item written before [17:39] the Binance hack where I'm waiting for [17:41] DPRK's favorite time of the year to [17:44] break things, which is winter holidays [17:46] when people are distracted. So that [17:48] already happened. That's that's a that's [17:50] a thing. [17:52] Um we're going to see multiple 100 [17:54] million plus chain chain killer events. [17:58] 95 plus compromises a month, [18:01] overwhelming. I have to track every [18:03] single one of them every week, so it's [18:05] it's brutal. [18:07] DeFi projects will more and more so will [18:10] be compromised through their centralized [18:12] dependencies. [18:14] And the last but the one that worries me [18:16] so much is the unholy birth of gray [18:19] market bounty brokers that already [18:21] exists in traditional security, [18:24] where vulnerabilities are not reported [18:26] to the projects, but in a reported to a [18:28] gray market where they're sold to the [18:30] highest bidder who are not necessarily [18:33] the the projects themselves. [18:36] We don't want to see that here. [18:41] So is there hope for redemption to all [18:43] of this? [18:45] I think so. [18:47] All the things that I'm going to share [18:48] with you, they already exist. They've [18:49] been tried this year and the past year [18:51] and they work. We just have to choose to [18:54] adopt them. [18:58] On AI and defense, [19:02] if you are a defender and your idea for [19:04] what you're going to do this year and [19:06] next year is to just throw AI at what [19:08] you're already doing, you will lose. [19:11] You will fall behind the race. [19:14] Instead, I want you to step back and [19:17] realize that as a defender, you are the [19:19] ones that are setting the terrain for [19:22] the attackers and AI and their bots or [19:24] whatever to operate it. [19:27] So why can't you design a terrain [19:30] that is so difficult to operate in, so [19:32] difficult to navigate that it drains [19:35] even the most well-funded attackers? How [19:38] can you burn their tokens to the ground? [19:42] And before we even start talking about [19:44] AI, I want to reintroduce some basics. [19:48] How can you lock down what you already [19:51] have? How can you set hard rules with [19:54] invariance, setting hard privileges, [19:57] allow lists? [19:59] If you can't lock it down, cap it. [20:02] There is no reason, I've been saying [20:03] that for years, for your vault to be [20:05] drained in a single transaction. [20:09] Set the outflow limits, time locks, [20:11] multisigs. Last but not least, add a [20:15] human element, which is the most [20:17] difficult part for automation to bypass. [20:21] Cheapen your mistakes. Realize that you [20:23] will mess up and that's okay. [20:25] So what can you do [20:27] to make your mistakes not existential. [20:31] Set your guardian accounts that will [20:32] save you at the last moment. [20:34] Set automatic circuit breakers if you [20:36] are if you are an exchange and one of [20:39] your chains was already drained maybe a [20:41] good idea to automatically shut down [20:43] everything else or drain it. [20:47] Last but not least [20:49] with the Black October especially [20:51] just operate under the assumption that [20:53] you are already hacked. Your laptop is [20:56] hacked. [20:57] Your developer is already hacked. [21:00] So how can we build a resilient security [21:03] program that can survive not one or two [21:06] but more compromises of your machines, [21:08] of your keys [21:10] and still survive. [21:13] Then and only then can we start talking [21:15] about AI. And I'm not talking about [21:18] saying [21:19] Claude hack me do not do not fail. [21:22] That's not going to work. [21:24] Build a threat model. No one else [21:26] understand [21:27] the weak points and the most critical [21:29] parts of your protocol and your project [21:32] than you. [21:33] And once you identify those sensitive [21:36] points then apply AI to it. [21:39] Run models 24/7 against it. When a new [21:42] model gets released update your harness, [21:45] run it again. Do that before someone [21:48] else does it for you. [21:52] And speaking of models [21:55] a reminder from the hugging face [21:56] interface you cannot rely on commercial [21:59] models for security work. Even the most [22:02] expensive and permissive models that you [22:04] can obtain today will still have guard [22:07] rails. Worse those guard rails may be [22:10] applied to you at a moment's notice just [22:12] at the time when you most need it in the [22:14] middle of the incident. [22:16] You must practice self-reliance. Either [22:18] run your own infrastructure or rent a [22:21] redundant one. Practice with it, build [22:23] the harnesses so you know exactly what [22:25] do to analyze the locks, to reverse [22:27] engineer contracts. [22:29] Have that infrastructure on the ready, [22:31] so when the worst hits, [22:34] you will be ready for it. [22:39] And of course, what do we do about North [22:41] Korea? [22:42] Well, for that, I invite you to look [22:45] back in history at traditional finance [22:48] that we so like to make fun of. [22:50] The thing is, exactly 10 years ago, they [22:53] were in the exact same spot we are right [22:56] now. [22:57] In 2016, North Korea attempted to steal [23:00] $1 billion from the Bank of Bangladesh. [23:03] How many of you are familiar with that [23:05] incident? [23:06] Right? [23:08] The hack was unsuccessful by sheer luck. [23:12] They made a typo. They made a typo in [23:14] the Swift transfer form. Someone in the [23:17] Federal Reserve Bank of New York caught [23:18] it. Great. [23:20] There were a few more attempts following [23:22] the hack. [23:23] So what happened next? [23:25] The Swift network sat down all of the [23:28] member banks [23:29] and instituted a stringent security [23:32] controls that all member banks must [23:35] apply. [23:36] Secure enclaves, [23:38] 24/7 hotline and incident response, [23:42] shared threat intelligence. [23:45] And what happened within 4 years? There [23:47] were a few more attempts, but by 2020, [23:50] North Korea had no more successful hits [23:53] against traditional banking [23:54] infrastructure where they actually stole [23:56] some assets. [23:59] And by then, North Korea moved on to a [24:01] much [24:02] easier target, which is us. [24:06] So can we have our very own Swift [24:09] moment? [24:11] Can we finally decide to self-regulate [24:15] before the regulation is forced upon us? [24:19] If you saw the FBI notice this morning [24:23] telling various projects, [24:25] such as Store Chain, such as node [24:28] to do something about funds. [24:31] That was your first warning, that [24:33] something is coming. [24:35] And I have a feeling that this industry [24:36] and people in this room much prefer to [24:39] self-regulate instead of someone telling [24:40] them what to do. [24:42] So can we adopt an industry-wide [24:45] security controls? Think seal [24:48] certificates, [24:50] but with teeth, with incentives. [24:53] Apply security controls, secure [24:56] yourself, [24:57] and enable us to invest in you, to [24:59] integrate with you, to provide liquidity [25:01] and insurance, or stay away in your own [25:04] walled garden. [25:07] Can we agree on the security first [25:10] neutrality? [25:12] I get it, we want to be decentralized, [25:15] but that is not an excuse for letting [25:17] stolen funds and abuse of your platform. [25:21] All you have to do is to publish [25:24] an intervention policy, define the exact [25:27] times and boundaries when you will apply [25:29] it, [25:30] be accountable to it and set realistic [25:32] timelines measured in minutes to hours, [25:35] not days or weeks. [25:38] Last but not least, shared intelligence [25:40] and coordination is critical. [25:43] Again, I'll keep on using seal because [25:45] this is this is our best hope right now. [25:48] We have seal 911 and seal intelligence, [25:50] but these are perpetually underfunded, [25:53] overwhelmed projects. How can we support [25:56] this and more [25:58] in our ecosystem? [26:02] We must bring security researchers back. [26:06] And it's not hard. It's not some magic [26:08] trickery. We just need to align the [26:10] incentives and make it worth their time, [26:13] not just financially, but even morally, [26:15] not treat them like they're doing [26:16] something criminal by telling us of the [26:18] vulnerabilities we're trying to fix. [26:24] Last but not least, [26:26] to all of you security researchers, [26:28] defenders, [26:30] bounty hunters, [26:31] who keep again and again and again that [26:34] AI is going to replace us. [26:37] It is not. [26:40] AI is an excellent tool, but that's all [26:43] it is, a tool in the hands of security [26:45] researchers. [26:47] Even if the whole world is trying to [26:49] tell you otherwise, and even sometimes [26:51] you yourselves, it is not a replacement [26:53] for human creativity, [26:55] for the grit and desire to learn and [26:59] explore. [27:01] And if you need any evidence of that, [27:03] just look at the talks at this very [27:05] conference written by the creative [27:09] minds [27:10] that are exploring this world together. [27:17] Folks, [27:19] it's always darkest before the dawn. [27:23] And it was a very dark year, the darkest [27:26] I've seen in the eight years I've been [27:27] following the space. [27:29] But I am optimistic [27:32] that we will bring the light [27:34] to our industry once more. [27:39] We have the tools, [27:42] the technology, and most importantly, [27:44] the people. They're all already here. We [27:46] just need to choose to use them, [27:49] choose to win. [27:51] We can choose to bring security [27:52] researchers back. [27:54] We can choose to use the same offensive [27:57] tech as the best defensive tools we've [28:00] ever had. [28:02] We can choose to unite and protect this [28:05] ecosystem together instead of protecting [28:08] our little kingdoms. [28:11] So let's light this fire together [28:14] because this time around the power to do [28:17] so is in our hands. [28:20] Thank you.