DSS 2026 / Day 1 / Amber Hall
Auditing Is Not Bug Hunting: Securing Systems Through Risk Management
0xa5df
Source: YouTube automatic caption track · Exported 2026-11-16. Timestamps link to the source video. Cue end times were not supplied.
Transcript
0:01Hi, good afternoon.
0:04Um, I'm going to talk today about our
0:06responsibility as auditors and how it
0:09goes beyond just finding bugs.
0:12Um, because when I started working at at
0:15doing the bug hunting and participating
0:17in contest, my goal was to find as many
0:21bugs as I can in order to maximize my
0:23payout. But I didn't have a
0:25responsibility for the system as a
0:26whole. If I lost if I missed a bug, it
0:29was a lost opportunity, not something
0:32that that I should be too much concerned
0:33about.
0:35And then I joined Storra and I started
0:38to do client audits. And now I started
0:40to have a responsibility for the system
0:41as a whole. I had to make sure that I
0:44don't miss any bugs and especially those
0:47who lead to
0:50um who lead to, you know, losses of
0:53millions of dollars.
0:57So how did that change my work? In two
1:00ways. First one is prioritizing. Now in
1:03bug bounties and contests as well, you
1:05have to prioritize by severity in order
1:08to maximize your payout.
1:10But I would argue that this uh severity
1:13is often subjective or theoretical and
1:15doesn't actually represent the actual
1:18potential damage uh of each bug.
1:22And when it comes to um actually
1:26protecting the system, you have to be
1:28dead practical about the potential
1:30damage of each bug.
1:33So usually I classify this into like
1:35minor issues which are not that
1:36significant, then significant issues
1:38which you don't want to miss. But then
1:40there are those uh issues that make it
1:42to the news with millions of dollars
1:44stolen. And this is the kind of stuff
1:46which you really really don't want to
1:47miss.
1:50Now the second way is
1:54by
1:56realizing that there are some ways to
1:57protect the system beyond just finding
1:59bugs and this is by introducing new
2:01hardening mechanism which would protect
2:04the system against some threats which
2:06are not due to onchain bugs. So a recent
2:08example is that Hedron um protocol which
2:13I audited a few months ago
2:15where
2:17they have a prop AMM. Now propriety AMM
2:20unlike a uh decentralized AMM where the
2:23price is determined by an algorithm.
2:25Here the price is determined by a single
2:28owner who owns the liquidity in the pool
2:30and also updates the price through an
2:33oracle service. Now this Oracle service
2:37is um you know it has up to like million
2:41updates per day. So you can't have like
2:43a human approving each update with the
2:45cold wallet. So it's a hot wallet which
2:47is stored on some server and this is
2:49obviously has a potential of getting
2:52compromised.
2:54Now what happens if it's compromised is
2:57an attacker can basically steal all of
3:00the funds in the pool. So they can
3:02basically set a price so that they can
3:03give a very fraction of uh soul for
3:06example for the entire USDC and then
3:08they give a fraction of use set a price
3:10to the other way and give a fraction of
3:12USDC for the remaining soul in the pool
3:15and this way in a single transaction you
3:17can just empty the pool.
3:21So what kind of mechanism did we
3:23introduce in order to defend against
3:25this kind of attack is the draw down
3:28guard. It's basically
3:30uh a mechanism which monitors that value
3:34of the liquidity of the pool and if it
3:36detects that the value is decreasing too
3:38fast then it pauses the pool and doesn't
3:40allow any further swap
3:43because a slow decrease can often happen
3:46naturally due to price changes. Um, but
3:49if we see that it's just moving too
3:51fast, like for example, 2% in 30
3:54seconds, then we should pause the pool
3:56and don't allow any further swap.
4:00Now, this slows down the attacker. It
4:02doesn't allow them to empty the pool in
4:04a single transaction,
4:06but it still allows them to do a slow
4:09drain over a few hours. Even with this
4:11mechanism, they can steal significant
4:14amount of the pool. So the second part
4:17of this mechanism is to monitor the pool
4:20through an external service. Uh a pa a a
4:24bot which has the pause a role which can
4:26only pause the the pool and not unpause
4:29it and in case it detects any invalid
4:31prices then it pauses the pool and the
4:35attacker can't um swap anymore. So those
4:38two mechanism together should
4:40significantly reduce the total amount of
4:43funds that an attacker can steal in case
4:45the uh oracle ro is compromised.
4:50Uh because you have only a few seconds
4:52before before the puzzle bot kicks in
4:54and the amount you can steal is should
4:56be you know depending on the
4:58configuration only a few percent um
5:02before the draw down guard kicks in.
5:06So yeah, so basically this is an example
5:08of how we can this is not a bug in the
5:10system. This is an additional mechanism
5:13uh which we give to the client to
5:16protect their system. And basically yeah
5:19this is this is the reason why the
5:20clients come to us. They come to us not
5:22just to find bugs but to protect the
5:24system and make sure they don't make it
5:26to the news with millions of dollars
5:27stolen. So this is what we should
5:29deliver to them. Thank you.
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: 5f7de7db1045b5f5d29fb02bb773434109e089f64ac3d4ee4338b148488dd087