DSS 2026 — Recovery Is the Attack Michael Lewellen Source video: https://www.youtube.com/watch?v=88xjsxH9QAQ Automatic captions — not human-reviewed. Complete spoken-audio coverage is unverified. Source language tag: en (English). Source export SHA256: 5b702f8c415240133df67873d4817ec3b37a43f0f8d2f33642ea5a0faf576b13 [0:01] All right. Hello everyone. [0:03] Today I'm talking about [0:05] how to harden your accounts and deal [0:07] with the fact that recovery itself can [0:09] be an exploit. So we will jump right in. [0:13] All right. [0:15] And why, you know, getting back into a [0:16] wallet and recovering it for a user can [0:19] also be the way that an attacker gets in [0:20] themselves. [0:23] So ideally every wallet should have a [0:25] recovery path. Recovery is good because [0:27] if people lose access to their wallet, [0:29] then they've lost all of their money. [0:30] For example, Bitcoin, which is the first [0:32] crypto wallet [0:34] effectively that came out in existence, [0:36] had a really rough you know, early few [0:39] years where a lot of people lost their [0:41] funds. Even if we put aside Satoshi, [0:43] you know, billions of dollars have [0:45] effectively been lost. We've all heard [0:46] the story of the guy digging his hard [0:49] drive out of a landfill, never to find [0:51] it because his girlfriend threw it out. [0:53] So these are the scenarios that we want [0:54] to prevent. We need to give consumers [0:56] good way to recover their funds. [0:58] However, on the flip side, if you make [0:59] it too easy to recover access to an [1:01] account, an attacker is going to use [1:03] that as their [1:05] mechanism for taking over the account [1:07] and stealing the funds. [1:08] So we need to find a middle ground here [1:11] and we're going to do so in the talk [1:13] just by going through like a lot of the [1:14] scenarios that we might deal with and [1:16] some principles that we could apply. [1:19] So first of all, there is some good [1:20] news, which is that login methods over [1:23] the last 10 to 20 years have gotten a [1:24] lot stronger. We've gotten far better [1:26] methods both within the cryptocurrency [1:28] industry, but also in the wider software [1:29] industry for having harder [1:31] authentication. [1:32] You know, we have cases where initially [1:33] we started out with username and [1:35] password, which I think a lot of the [1:36] people in the crowd will know are [1:38] fundamentally insecure. [1:39] You know, passwords can be stolen. [1:42] They're often reused. Even if you're [1:43] using a good password manager, [1:45] you know, it just takes one flaw or one [1:48] lucky guess by an attacker for [1:50] potentially the entire account to be [1:51] compromised. So then people came out [1:52] with SMS codes, which was an [1:54] improvement, but also [1:56] has its own problems with SIM swapping, [1:58] people being able to take over the [1:59] through [2:00] uh you know getting through the telecom [2:01] provider. Uh and then but we've [2:03] over time added better methods, [2:04] authenticator apps, uh which are much [2:06] more secure, pass keys, and then even [2:08] hardware keys uh that you could have. So [2:10] we have a lot better methods and if you [2:11] have multi-factor authentication [2:13] according to some uh uh research from [2:15] CISA, you know you're you're probably [2:17] seeing a 99% literal reduction in [2:19] account compromise risk. Um you know so [2:22] that's rough number but generally [2:23] speaking you get the idea. MFA is going [2:25] to be massively more secure. Um however [2:28] um if you have greatly secured the front [2:31] door, you now have to worry about the [2:32] back door. And if the back door doesn't [2:33] have the same number of locks, the [2:35] attacker is going to have an easier time [2:36] getting in nonetheless. Um so this is [2:38] something that actually the UK National [2:40] Cyber Security Research Center has has [2:42] come out and said uh is clearly a [2:44] pattern they're recognizing whenever [2:45] pass keys are introduced to accounts. Uh [2:47] potential attackers are now focusing on [2:49] the weaknesses around account recovery [2:50] and reset requests. You know how do you [2:52] reach out by email or through email [2:53] recovery, phone, maybe even reaching out [2:55] to support and finding you know the [2:57] amount of information or personal [2:58] information you have to impersonate [3:00] someone to try to recover the account in [3:01] that way uh since the pass key itself is [3:04] is very difficult and almost impossible [3:05] for the attackers now. Um so we now have [3:08] to deal with this this second metric and [3:09] we have to consider okay how do we [3:11] harden that recovery in such a way that [3:13] it is still useful for the user to [3:15] recover um but it's not going to be [3:17] itself the attack vector. [3:19] Um so there's three back doors that we [3:21] can think about for this um or rather [3:24] three example cases. Um the first one is [3:27] the the help desk with Coinbase where in [3:29] 2025 uh there was a massive incident [3:32] where a a subcontractor of Coinbase [3:34] doing support uh leaked a ton of [3:36] customer information, uh details that [3:38] allow them to run and reach out to those [3:40] customers and essentially convince them [3:42] uh to go through recovery method, [3:44] essentially had a lot of fake recovery [3:45] calls and led to a lot of customer funds [3:47] essentially being sent um you know un- [3:50] unknowingly to attacker wallets. So they [3:53] basically were told you're locked out, [3:54] they were told recover in this way, send [3:56] the money to these wallets, and then it [3:58] ended up actually being the attacker's [3:59] wallets and they stole the the So not [4:01] exactly a straightforward recovery story [4:02] but like a good example of how like [4:04] relying on something like a support desk [4:05] can actually lead to the accounts [4:08] themselves being compromised. Uh, and [4:09] that had a massive impact. There was, [4:11] you know, between roughly $180 and $400 [4:13] million that were likely lost uh to that [4:16] alone. Um, so it's a good case where [4:18] like also support desk having personal [4:19] information about customers uh can also [4:22] be a problem. So you have to be careful [4:23] how much you trust humans in that mix. [4:25] Um, another example of this is cloud [4:27] backups. A lot of people who have used [4:29] self-custodial wallets such as MetaMask [4:31] are familiar with the fact that it [4:32] usually gives you the option not just to [4:34] write down a seed phrase, which some [4:35] people maybe do insecurely because they, [4:37] you know, instead of writing it down, [4:38] they just copy paste it somewhere on [4:40] their computer and then their computer [4:41] gets compromised, they lose access to [4:43] their account. Um, MetaMask did [4:45] introduce the ability to essentially [4:46] have an encrypted version of their [4:48] backup that was placed in their iCloud [4:50] account. Um, however, this now means [4:53] that if the uh Apple account is [4:55] compromised, the backup could also be [4:57] exposed. Um, so this did happen in 2022. [5:00] There was a case of a user who had their [5:01] Apple ID phished through the normal [5:03] phishing methods that you might expect, [5:05] and through that uh the attacker was [5:06] able to gain access to their seed backup [5:08] in iCloud. Now, you might think, well, [5:10] if that is encrypted, that shouldn't be [5:12] a problem. However, you have to remember [5:13] these are passwords that users are [5:15] inputting into MetaMask every time, [5:16] which means they're probably uh maybe [5:18] creating a password that is not as [5:20] secure as it ideally should be. So the [5:21] hinge point is now on a user remembering [5:23] a password. And again, we talked about [5:25] email and password is not the most [5:27] secure method. So just having one pass [5:29] phrase the user has to remember or they [5:31] lose access to the wallet, which is also [5:33] something we don't want for recovery, is [5:35] now a problem. And so that led to the [5:37] fact that the password was easy enough [5:38] for a computer to randomly generate uh [5:41] and and guess um led to the wallet then [5:43] being drained and and losing 655k in one [5:46] specific instance, but there are other [5:48] cases out there where you can find [5:49] accounts being compromised, backups [5:51] being leaked, and the encryption not [5:53] being strong enough because the password [5:54] had to be something the user remembered [5:56] uh and and that led to a problem. And [5:57] you know, then if you're like, well, you [5:59] know, use a password manager or [6:00] something else, yes, but again, the more [6:01] complex the setup gets uh the harder it [6:04] is to get users to do the recovery [6:06] effectively and safely. [6:07] And then the last case was Guardian, uh, [6:10] a Guardian uh, essentially takeover [6:12] attempt with Loopring, which was a [6:13] wallet that was providing essentially a [6:16] kind of Guardian recovery method. The [6:17] idea being like if you lose access to [6:19] your wallet, uh, there is a way for [6:21] essentially Guardians to recover it for [6:23] you, um, which would be kind of like [6:25] trusted actors that could identify you. [6:27] Um, and Loopring itself had set itself [6:28] up as a default Guardian where you would [6:30] use essentially a two-factor [6:31] authentication uh, to authenticate the [6:33] Guardian to recover your account. Well, [6:35] that Guardian was then compromised. So [6:37] Loopring had a big security incident [6:39] where uh, essentially their Guardian [6:41] account was compromised. So it as a [6:43] recovery agent could then go to all [6:45] these wallets and effectively recover [6:46] them, uh, except recovering them for the [6:48] attacker, not the users. And that led to [6:50] $5 million being drained. Um, so these [6:52] are all like different cases, but cases [6:54] where recovery was the weak point and [6:56] led to a large loss of funds uh, using a [6:58] measure that was supposed to protect [6:59] user funds because it's trying to [7:00] prevent losses by forgetting passwords, [7:02] authenticators, etc., instead leading [7:05] attackers uh, in through the back door. [7:08] Um, so what do we do about this? Um, [7:10] what are what are some practical [7:11] considerations taking away from this [7:12] that we can actually implement? Um, [7:14] well, obviously like yeah, email only [7:16] and pass phrase backups are practical, [7:17] but they they can increase uh, you know, [7:20] make it easier effectively to attack. So [7:23] uh, and MFA helps, but you know, you [7:25] have to consider the fact that if you [7:26] make MFA the requirement for recovery, [7:30] um, they could make it difficult to [7:31] effect recovery if you lose too many [7:32] factors. Um, so you can see some [7:34] examples here of just like you know, [7:36] yeah, SMS reset, email code, support [7:38] request resets. These are easy for the [7:40] user, but they're also fairly easy for [7:42] the attacker as well. Um, you know, and [7:44] if you start to combine factors or you [7:46] go with things like paper backups, [7:47] again, assuming the user can uh, do the [7:50] paper backup effectively, uh, then that [7:53] is going to increase secure increase [7:55] security ideally, um, but can still lead [7:58] to some problems with the user actually [7:59] being able to implement them [8:00] effectively. So a few takeaways could be [8:03] you know, factors in series. So instead [8:04] of um, [8:06] you you just being essentially [8:07] encouraging the user to have many [8:08] different authenticators and then being [8:09] able to combine them in different ways. [8:11] Um that can improve and then so you have [8:13] factors of the threshold. You still have [8:14] MFA, but if you lose one of the factors [8:17] in MFA, uh you have other factors that [8:19] you can bring in. So instead of an and, [8:20] it's an or effectively. [8:22] Um and then uh also you can introduce [8:25] time delays. So if you are potentially [8:27] providing someone the ability to recover [8:29] uh using less factors than they normally [8:31] would, uh then you can introduce a [8:33] cooldown period along with a [8:34] notification capability that tells the [8:36] user, hey, someone's trying to recover [8:37] your account. It'll be recovered in, you [8:40] know, one day, two days, three days, [8:41] however much time you think you can [8:43] practically provide. Uh and that [8:45] potentially gives them a way to stop an [8:46] attacker that might be trying to use [8:47] that recovery method as an attack. [8:51] So what's the general rule here? Well, [8:53] don't make your recovery uh you know uh [8:56] less you know basically don't make it [8:58] weaker than your login because then the [8:59] recovery method is going to be targeted [9:01] by attackers. Um and ideally make it as [9:03] strong uh as the login if not stronger [9:05] uh while still trying to preserve some [9:07] accessibility for the user. Um so you [9:09] don't want a single point of recovery. [9:10] Ideally there's no like just SMS email [9:12] only fallback. Uh it still requires some [9:14] sort of multi-factor authentication to [9:16] recover. Um you should assume backups [9:18] can leak, so you need to make sure that [9:20] they're secured with high entropy [9:21] secrets or there's some second factor to [9:24] decrypt. Um you should ideally have some [9:27] step-up uh requirements. So if someone's [9:28] about to do like a change of their auth, [9:30] they should have to uh maybe do some [9:32] step-up auth to do so. You want to [9:34] notify in every recovery event and maybe [9:36] give a time delay uh when you do that [9:38] recovery so the users have a chance to [9:39] catch it. [9:41] Um so one other closure is, you know, [9:43] I'm working turnkey. These are things [9:44] we're trying to focus on there. Um these [9:45] are principles you can apply to any [9:47] setup, but we're trying to specifically [9:49] focus on making it easy to have multiple [9:51] authentication methods for a user uh [9:53] using secure enclave so the key material [9:55] itself is never exposed and then having [9:57] an API that's fairly flexible so [9:58] developers can implement this without [9:59] being security experts themselves. [10:02] Um so yeah, generally speaking. And one [10:04] thing I will say like as a maybe nod to [10:06] some of the people that we're also [10:07] competing with is I like the Fireblocks [10:09] and Fortify approach where they actually [10:10] send backup methods to users, but they [10:12] use seed phrases specifically as the [10:14] encryption method, and you can even have [10:15] threshold-based decryption of the [10:17] backup. So you could say a two out of [10:19] three seed phrase, or everyone has a [10:21] different seed phrase, and then from [10:22] there two out of those three people with [10:23] their individual seed phrases can then [10:25] decrypt the backup, which I I like uh as [10:27] a method that is an independent recovery [10:29] method, doesn't rely on a provider, uh [10:30] but still has pretty strong security [10:32] properties and and has a good chance of [10:33] a successful recovery. [10:35] Uh so that's it. Happy to talk more [10:36] after the talk.