DSS 2026 / Day 2 / Amber Hall
Top hacks since Devcon 7 - what did we learn?
Matthias Egli
Source: YouTube automatic caption track · Exported 2026-11-16. Timestamps link to the source video. Cue end times were not supplied.
Transcript
0:01Thank you. So welcome everyone.
0:04Um let's see. Yes. So this is the first
0:08talk of today. I'm happy to give
0:10this one to start of the day giving a
0:13rundown of the top hacks we had since
0:17like last Devcon, which actually is a
0:19long time ago. So it's a really huge
0:22list.
0:23Let's see how far we get.
0:25Quickly
0:27about myself. So I'm Matthias. I'm from
0:29Chain Security. We are one of the audit
0:32companies reviewing a lot of the
0:34different mostly complex DeFi projects.
0:38We also do a lot of research
0:40and and also like to present it.
0:42So
0:44among those things you might have had a
0:46talk yesterday on like smart contract
0:49quirks
0:50by Lotaire.
0:52There is another one right after by
0:54Doris
0:56also on tricky things around Solidity
0:58and Dirty Bytes.
1:01There is
1:02one at Devcon by Dominic and then
1:06I see this
1:08animations don't work. There is one
1:11at the end like kind of closing this day
1:14on
1:16a new kind of vulnerability found which
1:20gets elevated by Glamsterdam. So this is
1:23something I can highly recommend all of
1:24you to check out on.
1:27There there will be a bunch of talks on
1:28this on Devcon too, but the first talk
1:31will be here this evening by Janis.
1:35But let's get started.
1:37So
1:38the first hack
1:41I want to talk about is Bitget.
1:44Bitget
1:46is a centralized exchange
1:48and
1:50they got
1:51compromised for 388 million.
1:56So
1:58how did this
1:59The attackers here, they
2:03compromised
2:05initially a security appliance
2:09in their
2:11investigative report which was published
2:13it's just mentioned as product A.
2:16So we don't know exactly
2:18what kind of security appliance it is.
2:21There was also mention of a management
2:23software
2:24referred to as project B
2:27where we also don't know exactly what it
2:28is.
2:31But what happened here is that these
2:36initial compromises allowed the attacker
2:37to move towards the signing
2:40infrastructure of the hot wallets.
2:43And
2:45at that point of course
2:47you don't need a human to make an error
2:49anymore because you can make the system
2:52sign off on those transactions which you
2:55want to do. So
2:58the
3:00story here is that like what what did we
3:04learn basically from Bitget at this
3:07stage right? So the private keys were
3:10very well protected inside an HSM and so
3:13on right? So
3:14industry best practice you could say but
3:18the additional protection software they
3:20put in place to prevent hacks was
3:23actually in this case the one which
3:25allowed the attackers initially to
3:27enter.
3:29We can only speculate on why didn't they
3:33say which software they used? Was it
3:37some NDA and they couldn't say it's that
3:39and that vendor
3:41or was it something where
3:44they say like yeah it's
3:46it might compromise others if they'd
3:48known that the system is vulnerable.
3:52So sadly from some of these attacks we
3:55don't learn enough to really be able to
3:57protect others.
3:59The main thing we can take away is
4:02any
4:04piece of software, any component you put
4:06in your stack
4:08on the way to signing is potentially one
4:12which is vulnerable including your EDR,
4:17including your like web application
4:21firewall, stuff like this, right? So um
4:26in this case
4:29what else can we do to limit the damage,
4:31right? It's like limiting
4:33the
4:35single transaction damage, right?
4:37Basically, how much is in a hot wallet?
4:39Because in these cases, sometimes you
4:42have circuit breakers in place but then
4:45the question is how much damage is
4:47already done?
4:48Um
4:50what was special also here, it took
4:53really quite some time until they
4:55realized that their hot wallet signing
4:58infrastructure is the problem because
5:01suddenly things move, right? And you
5:02don't know
5:04has someone the keys? Is something else
5:06compromised? What's going on, right? And
5:08it took them quite some time actually
5:10to shut down their hot wallet
5:14uh signing infrastructure. So also this
5:16is like for emergency
5:18actions that you realize, okay, what
5:22what do you need to do to pause your
5:25system holistically? They very quickly
5:27paused actually withdrawals and deposits
5:31as the usual place where they were
5:33expecting to get hacked, but not the
5:35internal system which signed off on hot
5:37wallet uh um
5:39transactions.
5:42Now, this was some time ago already um
5:46but of course, we are here also because
5:49of a very recent one which is very
5:51similar, which is the Binance hack.
5:54So Binance as I guess all of you know
5:57sitting in here was compromised just
6:00four days ago by now
6:02um and for for a record-breaking
6:05um
6:07of more than six billion
6:11USD equivalent. Half of that money was
6:15recoverable by freezing the stable coins
6:18USDT and USDC mostly and a few other
6:21real world assets.
6:24First a big disclaimer, right? I mean
6:26what I'm presenting here is
6:28based on the few things we already know.
6:32Binance communicates of what's going on,
6:34focuses currently on enabling
6:37withdrawals again and and has seen a lot
6:40of
6:41like action in this regard. So
6:44luckily
6:46we could say it's one of the very rich
6:48exchanges which can
6:50as it looks like afford to
6:54use its treasury to compensate for this
6:56loss.
6:58So what happened here, what we know
7:00already
7:01a laptop allegedly, right? But this is
7:04really like preliminary information,
7:06right? And this is something I I want
7:07you
7:09to know that while Binance said this
7:11it's in the middle of an investigation.
7:14We will get to know if they share a lot
7:17more and all of this might be wrong
7:19after they realized what really
7:21happened, right? So but what they said
7:23so far is DevOps laptop got compromised.
7:26They saw some other compromises across
7:28their network. In the end
7:30a service container as they describe it
7:33was compromised this way which was
7:35allowed to sign
7:38and
7:40usually the way this works is that you
7:41have some container which has some
7:44privileged access to the HSM which is
7:46trusted, usually some certificates on it
7:48which can talk to the HSM, right? And if
7:50you get access to that container you can
7:51sign
7:53the the HSM trusts you to sign.
7:57What then happened next is we saw around
7:59like
8:01for past eight the first outflows.
8:04Actually Ripple was one of the first
8:06chains they emptied 15 minutes later
8:09roughly the very big transactions
8:11against Bitcoin and Ethereum happened
8:15here
8:18for Ethereum for Binance 14 their big
8:22hot wallet they replaced the
8:26wallet with a like
8:29smart wallet. I'll quickly go into that
8:31afterwards. They swept that across all
8:34the tokens
8:36and then actually the sweeper stayed
8:37active for three more hours and kept
8:41sweeping because there were incoming
8:43transactions actually during that time
8:44which then obviously also got drained
8:48and
8:49some of them were also like refilling
8:52because an automated system saw that the
8:54wallet the hot wallet was empty so it
8:56needed to be refilled, right?
9:02What did we learn here,
9:04right? So
9:06first I'd say we didn't really
9:08we will learn from this, right? This is
9:11not something where we can say we did
9:13learn anything yet. It's just happening.
9:16It's a very recent incident, okay? It's
9:18one of the largest right hacks
9:21ever perhaps I mean you probably can say
9:23it's the largest hack ever, right? Which
9:25really was able to pull off. There is
9:27that one the one that got away which is
9:29also very interesting case but this one
9:34I think will have impact beyond
9:39our bubble, right? This is one where
9:43people notice Bitget was very big of
9:45course it was very big but
9:48Bybit we will talk about soon too very
9:51big hack, right? But no one knew Bybit
9:53but Binance is almost equivalent with
9:56like centralized exchanges
10:00for
10:01for cryptocurrencies, right? So we can
10:04expect really to see some like
10:08real world impact. Regulation is one of
10:11the big things,
10:14Contagion through the ecosystem seems to
10:17be very limited currently, right? But
10:20it's one of these puzzle pieces like
10:22which fell to some degree. Binance
10:24doesn't now have
10:26the ability to react as freely anymore
10:29because they need to invest all that
10:31money to recompensate users. So
10:34one of our buffers is gone with that,
10:36right?
10:38We'll learn more what can make this
10:40attack harder.
10:41Very preliminary things here. What did
10:43we see like again,
10:46um,
10:47it took quite some time until, for
10:49example, other hot wallets they they
10:52plugged in something to sweep them to
10:54cold wallets because in this fog of war
10:57you don't really know what's going on.
10:59And if you look into how they swept the
11:01other hot wallets afterwards, you saw
11:03that it's a manual thing. It's like
11:06block after block. It took some time and
11:08so on, right? So so someone reacted here
11:11to some degree, I'd say manually.
11:13Um,
11:14the of course the size of a hot wallet
11:17is something we can discuss. Faster
11:20freezing across, um, the stable coins is
11:23also something we can discuss here,
11:25right? We will learn more. All of this
11:27is very preliminary.
11:31So quickly for you, because this is
11:33actually important, if you ever also get
11:37fished or so, what will happen to your
11:39wallet is you will get a sweeper on your
11:42wallet. This is what happens today. The
11:44same happened to Binance here on their
11:46Ethereum wallet.
11:48It will get it will get replaced with an
11:51EIP 7702,
11:53um, which which is a very efficient way
11:56to to interact with Ethereum, right? So
11:58uh, I want also emphasize it's not like
12:01the fault of 7702 in this case, right?
12:04And and the other hot wallet gets
12:05drained traditionally with normal
12:07transactions.
12:09But if you get hacked, you will get a
12:12A sweeper is
12:14um a relatively simple smart contract.
12:16What it does,
12:18when you send in ETH, it will
12:19immediately send it somewhere else.
12:21Um when you uh
12:24approve um
12:26or if if you have ERC 20s deposited,
12:30um it will allow the attacker to
12:32withdraw them even though you still have
12:35the private keys and the attackers
12:36doesn't, right? So they replace with one
12:39transaction the code to gain control.
12:41You can regain control like Binance did
12:44a few hours later, right? But until you
12:46do,
12:48everything you deposit will be swept
12:50away to the attackers.
12:52Behaving very similar as if your private
12:56key was compromised.
12:59To you, you will not notice. The small
13:01example I have here is kind of almost
13:04self-explanatory code-wise, right? But
13:07um
13:08the real ones we can study also the
13:12sweeper wallets out there. They are
13:16well engineered. Um of course people
13:18just buy them on the black market,
13:20right? So it's not something they need
13:21to develop for themselves, but it's
13:23really this um industry where you
13:26actually even give part of the profits
13:28to those who develop that and so on.
13:32Um
13:35so summarizing the very first attack
13:37vector here is like we have
13:41the
13:43hot wallet infrastructure of centralized
13:46exchanges
13:47and they are being
13:49attacked, very targeted attacks, and
13:51then these
13:53hot wallet infrastructure is being
13:54compromised.
13:57Um
13:58let's talk about an old one by now,
14:01right? But because it was after Defcon,
14:03it's still here. Um
14:05Bybit. Bybit was hacked for uh
14:091,430
14:11million. Um
14:13and this was a case where a supply chain
14:16vulnerability
14:18fundamentally was the issue. So,
14:22and the attacker managed to get to a
14:27developer of Safe,
14:29which was used by Bybit,
14:32managed to compromise that laptop and
14:36get
14:37some AWS keys which allowed them to
14:41deploy a JavaScript version, like a
14:44small change in the JavaScript of the
14:47Safe
14:48centrally hosted
14:50UI, which then in the end caused them
14:55caused Bybit to get hacked. And and the
14:57way this was done like you you had this
15:01Bybit multisig
15:03signing off on a regular transfer which
15:06looked benign, but behind this curtain
15:09what really was signed off was to change
15:12the master copy of that Safe to an
15:13attacker controlled one. And with that
15:17they obviously had control over the this
15:20treasury
15:21and were able to
15:23extract all the money.
15:25So, this is one of the cases.
15:28It's also some time ago where we learned
15:30a lot.
15:31And you will see a lot of projects out
15:34there
15:35which try to help
15:37to validate that when you sign on a Safe
15:41that you have different means to check
15:44what you are doing.
15:46I talked to
15:48few people here and what what they
15:49actually are often doing now is build
15:51their own thing there, right? I mean
15:54have Claude build their own thing, but
15:58this is something where we really
15:59learned
16:01that wait, let me get to the
16:03learning step that we can improve things
16:08by validating with independent
16:11ways, right? And that that we can
16:13develop that software and it's actually
16:14easier today than ever, right?
16:17So,
16:18learnings remove single point of
16:20failures, be very
16:23holistic about this, think deeply about
16:25where are your single point of failure.
16:27Limit again, limit the damage which can
16:31be done with one transaction, which to
16:33some degree you can also think of a
16:34single point of failure, one
16:35transaction, right? Then
16:38specifically here, provide multiple
16:40front ends. If you provide something
16:42like a service which others rely on,
16:44then ideally you'd give them some
16:46independent way which is harder for
16:49attackers to
16:52compromise at the same time as your main
16:55way.
16:58Then
17:00start to
17:02like
17:03put some pressure on to incentivize
17:06people to prove to you that they did
17:09this, right? So demand custodians to
17:11prove and
17:14we like Bitget, Binance and so on. Of
17:17course they had ISO 27001, of course
17:20they had SOC 2, right?
17:22But of course these things don't focus
17:24on the attack vectors we care about,
17:26right? They are for different measures,
17:29right? So they're still important, it
17:31still makes sense to get them, but it's
17:34far from sufficient.
17:38A lot of this also was actually what we
17:41talked about
17:43two years ago in Mumbai
17:46with like also a list of all the things
17:48you should think about
17:50at a presentation at Devcon and we'll
17:52have a like the same presentation
17:54actually two years again in a few days
17:57at Devcon again together again with
18:00muted from Polygon and with Nick from
18:02Safe.
18:04I'm skipping a few. Drift protocol, a
18:07perp dex was hacked for almost 300
18:10million.
18:11This is much smaller here than in my
18:14version, sorry.
18:16So just very quickly what happened here
18:19like they compromised Sinus
18:21by actually approaching them at a
18:23conference
18:25for half a year pretending to be like a
18:30what was it like a
18:32I think a market maker.
18:35They managed to then by compromising two
18:39keys or basically gaining the the
18:43access to the laptops where where those
18:45two keys were on to become protocol
18:47admins.
18:49They
18:50added fake collateral to this perp decks
18:54and they borrowed against it and and
18:56then they bridged quickly back to
18:58Ethereum.
19:01So this is a little bit similar to one
19:04of the incidents which really
19:07caused
19:08really big problems across the industry,
19:11right? So which is Kelp DAO or layer
19:13zero getting compromised for almost 300
19:16million.
19:18So
19:19roughly half a year ago they got
19:21compromised. The
19:24attack here was
19:26a pretty good one from like
19:30engineering perspective, right? What
19:31they had to do to compromise layer zero
19:34was impressive.
19:36So
19:38first social engineering again as so
19:40often to get
19:44some access to layer zero to to again a
19:47developer machine to the Terraform
19:50project
19:52access the cloud infrastructure,
19:54modified two of layer zero's internal
19:57RPC nodes but just in memory to trust
20:02fake blockchain data, right? Because
20:03layer zero needs to listen to what's
20:05happens on the blockchain and so on and
20:06react to it and then now those ones
20:09reacted to things which weren't real.
20:11But that's actually not enough, right?
20:13Because infrastructure was set up to
20:15need independent third-party validators.
20:18But in this case they DDoS'd the
20:22external party
20:24and then Zero fell back to say, ah, we
20:26still have a trusted one, which is us.
20:28In this case, also, of course, Kelp DAO
20:30had it configured that this was allowed
20:32and so on, but it was a risk back then
20:35which was like many did. It wasn't
20:38highlighted. Afterwards, of course,
20:40finger pointing is always easier and you
20:41should never do this and
20:44we had cases where where like people
20:46looked into this before Kelp DAO, which
20:49we reviewed and where it was
20:51highlighted. So it wasn't totally
20:52unknown that you shouldn't do this,
20:53right? But still it was widely done.
20:57Um, they
20:59um forged that uh
21:03bridge transaction basically and then
21:05there were like that attestation and in
21:07the end they were able to mint um
21:09unbacked RSEs.
21:12And the problem with that is that there
21:14was a lot like this was basically
21:16hardcoded one-to-one on Aave and allowed
21:18the attackers to liquidate two ETH by
21:22depositing to Aave and causing um and
21:26and other protocols, but Aave was the
21:27biggest one affected here with the
21:29biggest liquidity, causing
21:31what we now call DeFi United and so on,
21:34right? Requiring really a backstop, many
21:37people chiming in and so on to try to
21:40not have basically Aave having to um
21:45not be able to service like their debt
21:48anymore.
21:50And and we are still paying off that
21:51debt.
21:53Um,
21:56what did we learn here?
21:58There
21:59there were a lot of learnings on this.
22:01Like the obvious one is like have Layer
22:04Zero with multiple independent
22:05forcefully validator, but actually a lot
22:07of people are moving away to different
22:09infrastructures. Also those we'll have
22:10to see how well they hold up.
22:13Um,
22:14try better to isolate developer
22:16environment from production environment.
22:19This is a repeating thing across many of
22:21the incidents we have seen,
22:23but it's technically very hard and from
22:25we come, we build this stuff
22:28as developers and we grew as developers,
22:31right? So we tend to have a lot of
22:33access and power as developers.
22:36While the traditional industry kind of
22:39doesn't trust their devs, right? But but
22:41we are the devs who build this, so
22:42obviously we trust ourselves.
22:45Yeah, put more limits in place, the
22:47bridge outflows
22:50like safety guards of to which degree do
22:53you trust certain systems? We see this
22:56finally happening a little bit more and
22:58more, right? But
23:00the the limits in place in many of the
23:04decentralized exchanges and so on, they
23:06are none. It's like as much liquidity as
23:09there is, it's okay, right? And and that
23:11can be a lot.
23:14Um
23:16yeah, what else is there?
23:18Again, think holistically. All these
23:22pieces you have in your stack can be and
23:24will be attacked and many of these
23:26attacks show
23:27where different pieces have been
23:29attacked. So let's look
23:32quickly after this into some of the
23:34other hacks, right? So here attack
23:37vector was really
23:39the users signing malicious transaction.
23:44Um
23:46ledger ledger itself wasn't hacked, but
23:50what was hacked, right? Was their supply
23:53chain allowing someone
23:56to
23:58like put
24:00inside a ledger
24:02chip which would
24:04capture the seed phrase when it was
24:07generated and shown on the ledger and
24:10and basically phone home, right? Just
24:12send an
24:13a message via cellular network and and
24:16and that's it.
24:18So
24:20this likely happened over quite some
24:22time
24:24and
24:26over a few days then these funds were um
24:29like stolen until people realized what's
24:31really going on.
24:33Um
24:35here
24:36some pictures. Credit goes to Nick from
24:39Safe who provided those of how it looks
24:42like inside the ledger, right? So you
24:44have an additional board which shouldn't
24:47be there. You have this small battery in
24:49there to give it power. You have a a
24:52main board of course where it's
24:53connected to and you have the antenna.
24:55You can actually open your ledger and
24:56it's very easy to see, right? And I
24:58advise like if
25:01it really doesn't hurt to do this.
25:03Um
25:05what did we learn here?
25:08Uh
25:08a little bit tricky. Um I like try to
25:12limit your exposure when buying smart
25:14wallets, right? When buying when buying
25:15hardware wallets. Of course buy them
25:17directly from Ledger is not something I
25:20am here to say yes that is the right way
25:23because I
25:24myself decided not to do this directly
25:26due to their history of losing customer
25:29data when you buy directly from them
25:32which is also annoying, right? So it's a
25:34lot of trade-offs here. Um
25:38a better general trade-off here is try
25:40to have two or three different um
25:45hardware wallets for funds where it
25:47really matters, right? Obviously not for
25:48your like daily operations but for your
25:53savings
25:54having two independent one and a
25:56multi-sig behind it kind of easy
25:59nowadays and definitely advisable.
26:03Um
26:05cold card.
26:08Cold card
26:10similar size estimated a little bit hard
26:13to say how much it was how much it
26:15really was but estimated to be around
26:16100 million too.
26:19This
26:20a long time ago, 2021, right? They
26:24upgraded their firmware to a version
26:28where
26:29like
26:30they refactored they put the random
26:33number generator, which was um the one
26:35to generate the seed into a library.
26:37This was all well written, was all
26:39correct, but the library when it was
26:42actually built was linked to the
26:44software RNG instead of the hardware RNG
26:47of the cold card, causing weak
26:49randomness. And then many years later,
26:52attacker were able to actually brute
26:54force through the 32-bit which this seed
26:58generated to
27:01generate um the wallets, which
27:05ultimately at that point obviously held
27:08funds, right? And then were able to
27:10drain those.
27:11Um
27:14the
27:16thing here is like I mean the RNG inside
27:19a wallet is really
27:24the most most critical thing you can
27:26have. There's almost not and it's very
27:29hard to like
27:31if I think of a random number in my
27:33head, right? Can you prove that it was
27:34random? Of course not. It's also very
27:36hard to prove randomness, right? That's
27:37just like very hard. You can generate a
27:39lot, but
27:40uh you will not find out if the RNG was
27:4332 bits by by any external measure. The
27:45only thing you can do is internally look
27:47into it.
27:49What did we learn here? Um
27:54when you audit this more for the
27:56auditors, right? You need to take care
27:59not only of the code, but of the build
28:01because this was ultimately a build
28:02failure.
28:04Um combine independent sources of
28:06entropy. This would help like many
28:09things of defense in depth for your RNG
28:12would have made this harder.
28:16So attack vector three, we have people
28:19get your seed phrases,
28:21novel ways. We haven't seen these
28:23before, right? These are kind of new
28:25attacks.
28:27I'm skipping over Liquid Network,
28:29another 320 million hack here. TL;DR uh
28:34again a software bug caching um
28:39like LBTC
28:41is is like a yeah Bitcoin derivative and
28:44they minted that they were able to to
28:47circumvent that check but they were
28:49allegedly white hat hackers who returned
28:5280% and kept 20% as a bounty or whatever
28:55right
28:56um which which is not something we
28:59should support.
29:00Um
29:03Cetus
29:05let me see oh I have one minute left um
29:08so Sui's largest DEX was attacked with a
29:11nice vulnerability in that sense it was
29:14again a software bug right in in their
29:16system so um
29:18yeah
29:20um
29:22in a in a library one error right so I
29:25can quick no I will not walk you through
29:27this because ah sometimes it's big
29:29sometimes small okay it's kind of hard
29:31to read here.
29:32Um
29:35I'll skip through this due to time
29:37constraints how how the bug actually
29:40happened um as an example um perhaps the
29:45final vulnerable code you can see here
29:50the takeaway I have actually on this
29:53is like anyone who looks into this from
29:55you guys here to say like hey where's
29:57the error right
29:58you you might know then you will be fast
30:00if you don't know it will really take
30:03a lot of time until you figure out
30:05why this code is really broken.
30:09And a lot I mean
30:11a few minutes I guess for the best ones
30:14perhaps an hour right but uh still
30:16doable still doable
30:19today
30:20on the learnings when I tried now
30:24and I rewatched what um like I wrote
30:28about this before and I gave it to to AI
30:31to where's the bug then immediately it
30:32found the bug and they said I'll give
30:34the cheapest I'll take the weakest AI I
30:37can find still can find the bug
30:39immediately, right? So this one is easy.
30:42Time's up for me.
30:44Um so we'll skip balancer V2 which was
30:47the last one on the list.
30:49Uh yeah, thank you so much.
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: 03c44056cbcd6e5b36c43f1ca33d643dc9dcc161c67540a8f86e817b5c6f7d19