DSS 2026 / Day 2 / Amber Hall
Recovery Is the Attack
Michael Lewellen
Source: YouTube automatic caption track · Exported 2026-11-16. Timestamps link to the source video. Cue end times were not supplied.
Transcript
0:01All right. Hello everyone.
0:03Today I'm talking about
0:05how to harden your accounts and deal
0:07with the fact that recovery itself can
0:09be an exploit. So we will jump right in.
0:13All right.
0:15And why, you know, getting back into a
0:16wallet and recovering it for a user can
0:19also be the way that an attacker gets in
0:20themselves.
0:23So ideally every wallet should have a
0:25recovery path. Recovery is good because
0:27if people lose access to their wallet,
0:29then they've lost all of their money.
0:30For example, Bitcoin, which is the first
0:32crypto wallet
0:34effectively that came out in existence,
0:36had a really rough you know, early few
0:39years where a lot of people lost their
0:41funds. Even if we put aside Satoshi,
0:43you know, billions of dollars have
0:45effectively been lost. We've all heard
0:46the story of the guy digging his hard
0:49drive out of a landfill, never to find
0:51it because his girlfriend threw it out.
0:53So these are the scenarios that we want
0:54to prevent. We need to give consumers
0:56good way to recover their funds.
0:58However, on the flip side, if you make
0:59it too easy to recover access to an
1:01account, an attacker is going to use
1:03that as their
1:05mechanism for taking over the account
1:07and stealing the funds.
1:08So we need to find a middle ground here
1:11and we're going to do so in the talk
1:13just by going through like a lot of the
1:14scenarios that we might deal with and
1:16some principles that we could apply.
1:19So first of all, there is some good
1:20news, which is that login methods over
1:23the last 10 to 20 years have gotten a
1:24lot stronger. We've gotten far better
1:26methods both within the cryptocurrency
1:28industry, but also in the wider software
1:29industry for having harder
1:31authentication.
1:32You know, we have cases where initially
1:33we started out with username and
1:35password, which I think a lot of the
1:36people in the crowd will know are
1:38fundamentally insecure.
1:39You know, passwords can be stolen.
1:42They're often reused. Even if you're
1:43using a good password manager,
1:45you know, it just takes one flaw or one
1:48lucky guess by an attacker for
1:50potentially the entire account to be
1:51compromised. So then people came out
1:52with SMS codes, which was an
1:54improvement, but also
1:56has its own problems with SIM swapping,
1:58people being able to take over the
1:59through
2:00uh you know getting through the telecom
2:01provider. Uh and then but we've
2:03over time added better methods,
2:04authenticator apps, uh which are much
2:06more secure, pass keys, and then even
2:08hardware keys uh that you could have. So
2:10we have a lot better methods and if you
2:11have multi-factor authentication
2:13according to some uh uh research from
2:15CISA, you know you're you're probably
2:17seeing a 99% literal reduction in
2:19account compromise risk. Um you know so
2:22that's rough number but generally
2:23speaking you get the idea. MFA is going
2:25to be massively more secure. Um however
2:28um if you have greatly secured the front
2:31door, you now have to worry about the
2:32back door. And if the back door doesn't
2:33have the same number of locks, the
2:35attacker is going to have an easier time
2:36getting in nonetheless. Um so this is
2:38something that actually the UK National
2:40Cyber Security Research Center has has
2:42come out and said uh is clearly a
2:44pattern they're recognizing whenever
2:45pass keys are introduced to accounts. Uh
2:47potential attackers are now focusing on
2:49the weaknesses around account recovery
2:50and reset requests. You know how do you
2:52reach out by email or through email
2:53recovery, phone, maybe even reaching out
2:55to support and finding you know the
2:57amount of information or personal
2:58information you have to impersonate
3:00someone to try to recover the account in
3:01that way uh since the pass key itself is
3:04is very difficult and almost impossible
3:05for the attackers now. Um so we now have
3:08to deal with this this second metric and
3:09we have to consider okay how do we
3:11harden that recovery in such a way that
3:13it is still useful for the user to
3:15recover um but it's not going to be
3:17itself the attack vector.
3:19Um so there's three back doors that we
3:21can think about for this um or rather
3:24three example cases. Um the first one is
3:27the the help desk with Coinbase where in
3:292025 uh there was a massive incident
3:32where a a subcontractor of Coinbase
3:34doing support uh leaked a ton of
3:36customer information, uh details that
3:38allow them to run and reach out to those
3:40customers and essentially convince them
3:42uh to go through recovery method,
3:44essentially had a lot of fake recovery
3:45calls and led to a lot of customer funds
3:47essentially being sent um you know un-
3:50unknowingly to attacker wallets. So they
3:53basically were told you're locked out,
3:54they were told recover in this way, send
3:56the money to these wallets, and then it
3:58ended up actually being the attacker's
3:59wallets and they stole the the So not
4:01exactly a straightforward recovery story
4:02but like a good example of how like
4:04relying on something like a support desk
4:05can actually lead to the accounts
4:08themselves being compromised. Uh, and
4:09that had a massive impact. There was,
4:11you know, between roughly $180 and $400
4:13million that were likely lost uh to that
4:16alone. Um, so it's a good case where
4:18like also support desk having personal
4:19information about customers uh can also
4:22be a problem. So you have to be careful
4:23how much you trust humans in that mix.
4:25Um, another example of this is cloud
4:27backups. A lot of people who have used
4:29self-custodial wallets such as MetaMask
4:31are familiar with the fact that it
4:32usually gives you the option not just to
4:34write down a seed phrase, which some
4:35people maybe do insecurely because they,
4:37you know, instead of writing it down,
4:38they just copy paste it somewhere on
4:40their computer and then their computer
4:41gets compromised, they lose access to
4:43their account. Um, MetaMask did
4:45introduce the ability to essentially
4:46have an encrypted version of their
4:48backup that was placed in their iCloud
4:50account. Um, however, this now means
4:53that if the uh Apple account is
4:55compromised, the backup could also be
4:57exposed. Um, so this did happen in 2022.
5:00There was a case of a user who had their
5:01Apple ID phished through the normal
5:03phishing methods that you might expect,
5:05and through that uh the attacker was
5:06able to gain access to their seed backup
5:08in iCloud. Now, you might think, well,
5:10if that is encrypted, that shouldn't be
5:12a problem. However, you have to remember
5:13these are passwords that users are
5:15inputting into MetaMask every time,
5:16which means they're probably uh maybe
5:18creating a password that is not as
5:20secure as it ideally should be. So the
5:21hinge point is now on a user remembering
5:23a password. And again, we talked about
5:25email and password is not the most
5:27secure method. So just having one pass
5:29phrase the user has to remember or they
5:31lose access to the wallet, which is also
5:33something we don't want for recovery, is
5:35now a problem. And so that led to the
5:37fact that the password was easy enough
5:38for a computer to randomly generate uh
5:41and and guess um led to the wallet then
5:43being drained and and losing 655k in one
5:46specific instance, but there are other
5:48cases out there where you can find
5:49accounts being compromised, backups
5:51being leaked, and the encryption not
5:53being strong enough because the password
5:54had to be something the user remembered
5:56uh and and that led to a problem. And
5:57you know, then if you're like, well, you
5:59know, use a password manager or
6:00something else, yes, but again, the more
6:01complex the setup gets uh the harder it
6:04is to get users to do the recovery
6:06effectively and safely.
6:07And then the last case was Guardian, uh,
6:10a Guardian uh, essentially takeover
6:12attempt with Loopring, which was a
6:13wallet that was providing essentially a
6:16kind of Guardian recovery method. The
6:17idea being like if you lose access to
6:19your wallet, uh, there is a way for
6:21essentially Guardians to recover it for
6:23you, um, which would be kind of like
6:25trusted actors that could identify you.
6:27Um, and Loopring itself had set itself
6:28up as a default Guardian where you would
6:30use essentially a two-factor
6:31authentication uh, to authenticate the
6:33Guardian to recover your account. Well,
6:35that Guardian was then compromised. So
6:37Loopring had a big security incident
6:39where uh, essentially their Guardian
6:41account was compromised. So it as a
6:43recovery agent could then go to all
6:45these wallets and effectively recover
6:46them, uh, except recovering them for the
6:48attacker, not the users. And that led to
6:50$5 million being drained. Um, so these
6:52are all like different cases, but cases
6:54where recovery was the weak point and
6:56led to a large loss of funds uh, using a
6:58measure that was supposed to protect
6:59user funds because it's trying to
7:00prevent losses by forgetting passwords,
7:02authenticators, etc., instead leading
7:05attackers uh, in through the back door.
7:08Um, so what do we do about this? Um,
7:10what are what are some practical
7:11considerations taking away from this
7:12that we can actually implement? Um,
7:14well, obviously like yeah, email only
7:16and pass phrase backups are practical,
7:17but they they can increase uh, you know,
7:20make it easier effectively to attack. So
7:23uh, and MFA helps, but you know, you
7:25have to consider the fact that if you
7:26make MFA the requirement for recovery,
7:30um, they could make it difficult to
7:31effect recovery if you lose too many
7:32factors. Um, so you can see some
7:34examples here of just like you know,
7:36yeah, SMS reset, email code, support
7:38request resets. These are easy for the
7:40user, but they're also fairly easy for
7:42the attacker as well. Um, you know, and
7:44if you start to combine factors or you
7:46go with things like paper backups,
7:47again, assuming the user can uh, do the
7:50paper backup effectively, uh, then that
7:53is going to increase secure increase
7:55security ideally, um, but can still lead
7:58to some problems with the user actually
7:59being able to implement them
8:00effectively. So a few takeaways could be
8:03you know, factors in series. So instead
8:04of um,
8:06you you just being essentially
8:07encouraging the user to have many
8:08different authenticators and then being
8:09able to combine them in different ways.
8:11Um that can improve and then so you have
8:13factors of the threshold. You still have
8:14MFA, but if you lose one of the factors
8:17in MFA, uh you have other factors that
8:19you can bring in. So instead of an and,
8:20it's an or effectively.
8:22Um and then uh also you can introduce
8:25time delays. So if you are potentially
8:27providing someone the ability to recover
8:29uh using less factors than they normally
8:31would, uh then you can introduce a
8:33cooldown period along with a
8:34notification capability that tells the
8:36user, hey, someone's trying to recover
8:37your account. It'll be recovered in, you
8:40know, one day, two days, three days,
8:41however much time you think you can
8:43practically provide. Uh and that
8:45potentially gives them a way to stop an
8:46attacker that might be trying to use
8:47that recovery method as an attack.
8:51So what's the general rule here? Well,
8:53don't make your recovery uh you know uh
8:56less you know basically don't make it
8:58weaker than your login because then the
8:59recovery method is going to be targeted
9:01by attackers. Um and ideally make it as
9:03strong uh as the login if not stronger
9:05uh while still trying to preserve some
9:07accessibility for the user. Um so you
9:09don't want a single point of recovery.
9:10Ideally there's no like just SMS email
9:12only fallback. Uh it still requires some
9:14sort of multi-factor authentication to
9:16recover. Um you should assume backups
9:18can leak, so you need to make sure that
9:20they're secured with high entropy
9:21secrets or there's some second factor to
9:24decrypt. Um you should ideally have some
9:27step-up uh requirements. So if someone's
9:28about to do like a change of their auth,
9:30they should have to uh maybe do some
9:32step-up auth to do so. You want to
9:34notify in every recovery event and maybe
9:36give a time delay uh when you do that
9:38recovery so the users have a chance to
9:39catch it.
9:41Um so one other closure is, you know,
9:43I'm working turnkey. These are things
9:44we're trying to focus on there. Um these
9:45are principles you can apply to any
9:47setup, but we're trying to specifically
9:49focus on making it easy to have multiple
9:51authentication methods for a user uh
9:53using secure enclave so the key material
9:55itself is never exposed and then having
9:57an API that's fairly flexible so
9:58developers can implement this without
9:59being security experts themselves.
10:02Um so yeah, generally speaking. And one
10:04thing I will say like as a maybe nod to
10:06some of the people that we're also
10:07competing with is I like the Fireblocks
10:09and Fortify approach where they actually
10:10send backup methods to users, but they
10:12use seed phrases specifically as the
10:14encryption method, and you can even have
10:15threshold-based decryption of the
10:17backup. So you could say a two out of
10:19three seed phrase, or everyone has a
10:21different seed phrase, and then from
10:22there two out of those three people with
10:23their individual seed phrases can then
10:25decrypt the backup, which I I like uh as
10:27a method that is an independent recovery
10:29method, doesn't rely on a provider, uh
10:30but still has pretty strong security
10:32properties and and has a good chance of
10:33a successful recovery.
10:35Uh so that's it. Happy to talk more
10:36after the talk.
Source and accuracy details
- Names, organizations, technical terms, code and numbers may be misrecognized.
- Source wording and disfluencies are preserved; no editorial corrections were applied.
- Export contains whole-second cue starts only; cue ends and subsecond timing are unavailable.
- A full caption-track export does not establish that every spoken word is represented.
- Automatic YouTube captions; not human-verified against the audio.
Source export SHA256: 5b702f8c415240133df67873d4817ec3b37a43f0f8d2f33642ea5a0faf576b13