DSS 2026 — Your laptop is in scope too Jan Kalivoda Source video: https://www.youtube.com/watch?v=8Aapo-q1BSU Automatic captions — not human-reviewed. Complete spoken-audio coverage is unverified. Source language tag: en (English). Source export SHA256: c58dc59396f77611567ce0d81489071edd4a29198d742610814e87cf83302b61 [0:00] My talk is about personal security. So [0:04] when you are developing something or [0:06] auditing, so you don't get hacked during [0:08] that. This was initially supposed to be [0:12] a workshop, but [0:14] changed the talk, so it will be covered [0:16] just briefly. But definitely you can [0:19] then at your hotel apply all these [0:21] things, so you will be sure that you [0:23] will keep secure. [0:25] So why is this important? [0:28] You probably noticed also [0:31] few days ago the Binance hack. [0:33] So it was kind of severe and so from [0:37] what I heard it also somehow involved [0:39] compromised developer machine. [0:42] And these attacks are more and more [0:44] common. [0:46] You are you are now targeted anywhere, [0:48] so you should be [0:51] more focused. [0:54] Yeah. [0:56] Today we will cover four topics. [0:59] We will start with IDE, package [1:01] managers, GitHub, and your AI coding [1:03] tools. So the most important stack to [1:07] create or audit something I will say. [1:10] So starting with IDE, [1:12] you probably know that you have there in [1:14] VS Code or Cursor that JSON setting. [1:18] There are just [1:19] few of them. [1:20] First one is the workspace trust, which [1:23] is [1:24] recommended to be turned on because when [1:27] you open [1:28] some [1:30] some folder, [1:31] it will prevent execution within that. [1:34] So you just need to also click a button [1:38] to [1:39] trust it. [1:40] You can also turn off automatic tasks, [1:45] which is kind of recommended. Is there [1:47] anyone who is using [1:49] this task in VS Code? [1:53] Did not encounter anyone yet. [1:56] So [1:57] you should have this definitely [2:00] We will show later why. And then I [2:04] personally like [2:06] when when I when I have control over [2:09] updating some packages and essentially [2:12] all the software. So I'm just turning [2:14] off all the updates [2:16] and checking that always in my weekend [2:19] sessions manually [2:21] that I'm not pulling [2:23] any malicious version. [2:27] This is how the [2:29] workspace [2:30] window looks like. [2:32] And I'm not sure if it is clearly [2:35] visible, but this is example project [2:37] which can trigger arbitrary script on [2:40] your computer. [2:42] You basically open just some folder and [2:45] it has .vscode folder there and there is [2:49] defined a task. [2:50] So and this task [2:53] is executing harmless. [2:57] SH [2:58] file [2:59] and as you can see there is a lot of [3:03] other arguments that are passed [3:05] and those arguments are essentially [3:07] doing that it will be executed very [3:09] silently, so you won't notice that. [3:12] So [3:14] when you were there nothing happened and [3:17] when you click trust [3:19] then after 30 seconds, which is [3:22] the script, you got just executed it and [3:25] there there was created a file. [3:32] Then definitely [3:34] problem is with extensions. So [3:39] what I recommend for companies, you can [3:41] create [3:42] approved list of extensions where you [3:44] pin to exactly some version and you can [3:47] distribute it between all your employees [3:50] so you will ensure that no one will be [3:52] using any unapproved extension [3:56] in their editor. [3:58] And [3:59] reviewing your updates and also you may [4:01] be [4:03] someday in the past, you know, you have [4:05] VS Code probably installed for years, [4:07] and it's still evolving, it's having new [4:09] parameters and so on. You may be click [4:11] there that you have a full disk access [4:14] or other things. [4:15] So it's also good to review this. [4:17] Maybe you don't need it for that. [4:21] Then we have their package managers. [4:24] So [4:26] the flow basically for those basic [4:28] supply chain attacks is that there is [4:31] some maintainer of some package or his [4:33] CI gets compromised, [4:35] update is released, [4:37] and [4:40] you will just execute it on your machine [4:43] and [4:44] get for example credentials stolen. [4:52] So some basic settings what you should [4:54] keep in mind, [4:56] have scripts off. It's basically similar [4:58] to the thing we were describing in VS [5:01] Code. [5:03] And [5:05] also in VS Code, and I I'm not sure if [5:07] in VS Code itself, but I know that there [5:10] was that option [5:12] in VS Code. [5:13] You can set there a freshness gate for [5:17] some packages, and this should be also [5:20] definitely set [5:22] in your package manager setting. [5:25] Then [5:26] as always you have to check for [5:28] overrides and that your toolchain is [5:30] properly pinned. [5:33] And this is kind of funny because you [5:36] have a lot of package managers and they [5:40] are setting those things [5:42] pretty differently. [5:43] So let's say you will just check it, [5:47] set it correctly for npm, [5:49] let's say also for pnpm, [5:51] but then you will just change version [5:54] with nvm, let's say, and then you will [5:57] be using completely different config [5:59] setting, [6:00] which is not the one that you you were [6:02] supposed to use. So [6:05] again scripts can get executed. So, you [6:08] have to have it like covered uh [6:12] in all many possible ways [6:15] or you have to ensure that you will be [6:16] always using just the version you want [6:18] to and where you have the config [6:20] specified. [6:22] Uh this is example for uh [6:25] the gates. [6:27] Uh I personally like most PNPM. Uh I [6:31] think they have the best attitude to [6:34] security. So, I'm also enforcing it to [6:37] use it by my agents and always their [6:39] specific version range. [6:43] So, then we have a GitHub. [6:45] Uh [6:46] who is using GitHub CLI? [6:49] Here. [6:50] One? Just one person? I was expecting [6:52] more. [6:54] Well, um [6:55] I stopped using that a year ago because [6:59] I realized that by by default that tool [7:03] was saving [7:05] uh [7:06] your access key in plain text. And when [7:08] you removed the GitHub CLI tool, then uh [7:12] it's still there was the plain text [7:15] existing. [7:16] So, if you are using GitHub CLI, then I [7:18] urge you [7:20] to check this because you should not [7:22] have any secrets in plain text on your [7:24] machine. [7:26] And [7:28] then specifically for the GitHub, [7:32] uh [7:33] least privilege [7:35] approach is the viable way everywhere. [7:40] And also for Git, you have to be aware [7:43] of Git hooks. So, you should disable [7:46] them. [7:47] And more importantly, if you are for [7:49] example auditor and you will get a code [7:52] base from your client and you want to [7:54] push it to your let's say internal [7:56] organization to collaborate there with [7:58] your [7:59] colleagues, then you have to have keep [8:02] in mind to the actions because [8:05] there can be anything even though it is [8:07] your client, Uh [8:09] but you can potentially trigger some [8:11] malicious action. [8:13] So [8:14] what setting would I like [8:17] is Is it a laser? [8:20] No. [8:22] Is the last one [8:24] where you essentially approve all your [8:26] actions what you created. You are [8:28] creator, so you probably don't need to [8:31] be so much aware [8:32] if it's simple [8:34] or like that. [8:36] But [8:37] you can also select some specific [8:40] actions which you don't maintain and you [8:43] can pin them to specific hash [8:45] and this way you can just keep again [8:48] list similar like with extensions in VS [8:51] Code [8:53] of actions what what you [8:56] what to use. [8:59] And then definitely repository controls [9:05] are important to be set properly. [9:08] Then we have there section with AI [9:11] coding agents. [9:13] There I will have again a question. [9:16] Is there someone using auto approval [9:18] mode? [9:22] And someone clicking allow? [9:26] Okay. [9:27] I think [9:29] if you will be at this state like auto [9:31] approval if it will be configured good [9:35] it can be actually somehow useful, but I [9:38] think it will be still stopping you in [9:39] some bigger progress and I personally [9:42] don't trust it. [9:44] So when you're running agents I think [9:47] it's just a matter of time when you will [9:49] start with proper sandboxing. [9:52] So [9:55] I will recommend to start with [9:57] containers. [10:00] For example, [10:02] I'm not using Cloud Code or Codex. I [10:03] like Ollama Pi harness and it's by [10:06] default like YOLO. So [10:09] you can't even run it safely on your [10:11] machine. So this was also like my [10:13] motivation [10:15] to have it set properly. [10:17] But [10:19] even to even you are running it in [10:21] container or outside, [10:23] you can still [10:24] nicely configure these tool tools. [10:27] So let's say you can [10:30] create a recoverable cleanup. [10:32] You can replace [10:35] remove calls with trash command. [10:38] This can be little bit tricky because if [10:41] you are not cleaning your trash [10:42] regularly, then it will kind of grow. [10:47] Then you can have a specific [10:50] rules for feature branches and work with [10:53] Git. [10:54] This can be also stacked in the layers, [10:57] one layer on directly on GitHub and [10:59] another on the agent level. And [11:02] so you will ensure that [11:04] you have stacked up the security in [11:05] layers. [11:06] And [11:08] yeah, and so many other things. There [11:11] are examples how these configs can look [11:12] like. [11:15] You can deny some folders, do lots of [11:18] stuff. [11:20] For codex, it's in this way. [11:23] And [11:26] what I [11:28] would like you to do when you will be [11:31] coding with AI agents to have set the [11:34] sandbox. So for example, what works [11:37] nicely for me are just simple dev [11:39] containers when you can have for example [11:41] dev container for creating dev [11:42] containers and then you can create dev [11:44] containers for each task. [11:46] Mount always just the folder what you [11:48] need. [11:49] And [11:51] then you can run your agents and box in [11:54] this way while still they have access to [11:56] your laptop or if you are running it [11:59] remotely, that's even better when you [12:00] have a dedicated machine [12:03] for the coding. [12:07] Yep. And that's the talk. [12:11] Questions. [12:13] >> So when you're building um [12:16] that Docker image, how do you make sure [12:19] that the agent can escape and come back [12:22] to your, you know, laptop and actually [12:25] pull secrets off? [12:27] >> That's a very good question. Um [12:30] I'm not sure. Uh that's why still the [12:33] dedicated machine will be better. [12:35] But uh [12:36] uh I think [12:38] the versions of agents we are using uh [12:41] are much more constrained than than [12:42] those in some [12:44] lab scenarios where uh [12:47] where the frontier labs are are actually [12:50] testing them. So, they have them little [12:52] bit like unchained. Meanwhile, [12:55] we have still there lot of [12:58] guard rails and so on. So, I would [13:00] suppose that uh [13:02] like I think in security there is always [13:04] way [13:05] how you can hack yourself or someone [13:07] else. [13:08] But [13:09] uh it's generally about trying to do [13:12] most of the mitigation you can do and [13:14] still [13:15] preserve some comfortable way of being.