WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=JxNCuXkSWyQ NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/JxNCuXkSWyQ.html. A cue ends where the next begins, or 2 s after its last word. s1 00:00:12.640 --> 00:00:13.640 Hey everyone. s2 00:00:13.640 --> 00:00:15.080 Thanks for coming today. s3 00:00:15.080 --> 00:00:15.920 My name is Sagar. s4 00:00:15.920 --> 00:00:18.840 I'm CEO and co-founder of Speakeasy. s5 00:00:18.840 --> 00:00:24.560 We are an AI control plane that helps you securely scale AI within your business. s6 00:00:24.560 --> 00:00:32.560 So what this means is we we help take all the AI you have, put it on a governed path, and help you authorize all the actions it's taking. s7 00:00:32.560 --> 00:00:39.520 I'm I'm here today to talk to you about how we can actually intentionally authorize agent actions. s8 00:00:39.520 --> 00:00:50.040 I'm going to recreate a little incident that was in the news a couple months ago, use that to talk to kind of the scenario of how you can authorize actions and prevent prevent such an incident. s9 00:00:50.040 --> 00:00:55.320 The question of whether agents are capable at a task is long gone today. s10 00:00:55.320 --> 00:01:01.000 The real focus in most businesses is what happens when one of them does something. s11 00:01:01.000 --> 00:01:09.440 In most regulated industries, every action has to be has to be attributed to someone or something and and some identity. s12 00:01:09.440 --> 00:01:13.720 So that's really the the core of what I wanted to talk here. s13 00:01:13.720 --> 00:01:21.480 Let me first do a quick little animation I put together of the incident that I was talking about. s14 00:01:21.480 --> 00:01:35.000 If you remember back in April, I won't name names, but there was a startup founder at the startup who thought he was fixing a staging environment but accidentally deleted his production database. s15 00:01:35.000 --> 00:01:38.680 I've got this funnel animation to kind of remind you of that story. s16 00:01:38.680 --> 00:01:46.240 The summary of the story goes as I said, the developer asks a coding agent to fix a credential issue in staging. s17 00:01:46.240 --> 00:01:49.960 This is a pretty ordinary request, ordinary afternoon. s18 00:01:49.960 --> 00:01:56.640 The agent starts by reading the workspace and in an unrelated file, it finds a database access token. s19 00:01:56.640 --> 00:02:00.920 This is an infrastructure token with full access. s20 00:02:00.972 --> 00:02:05.692 It decides the fix is to delete the volume and recreate it. s21 00:02:05.692 --> 00:02:13.092 Except the volume it has access to is not staging and nothing is standing between that guess and the API. s22 00:02:13.092 --> 00:02:20.772 There's no identity on the call, no policy, no approval and really no record at all. s23 00:02:22.652 --> 00:02:29.332 So in one fell swoop we've deleted production, it's back up and all of that lived within the same network boundary. s24 00:02:29.332 --> 00:02:31.732 So it's all gone in one instant. s25 00:02:31.732 --> 00:02:38.172 Took about 10 seconds to kind of completely bring down the startups whole database and and their backup. s26 00:02:38.172 --> 00:02:49.132 Now what I want to leave you with as we kind of go on to the next part of this presentation is that the agent had valid credentials but it didn't have valid intent. s27 00:02:49.132 --> 00:02:55.932 That's the real difference between the kind of right and wrong in the situation. s28 00:02:56.532 --> 00:03:02.332 So my argument I'm going to make to you here is that the unchecked authority is the real failure. s29 00:03:02.332 --> 00:03:05.332 An agent can be wrong, it can miss context. s30 00:03:05.332 --> 00:03:09.252 It's pretty well understood now, you know, we've all been in that situation before. s31 00:03:09.252 --> 00:03:15.452 But what you didn't see in that scenario is that there was no exploit, there was no jailbreak, no prompt injection. s32 00:03:15.452 --> 00:03:17.052 There was really no malicious actor. s33 00:03:17.052 --> 00:03:23.212 There was nothing that anyone was doing intentionally to kind of harm the system or harm the company. s34 00:03:23.212 --> 00:03:26.932 The valid key did exactly what valid keys do, right? s35 00:03:26.932 --> 00:03:29.252 The agent did what it thought it should do. s36 00:03:29.252 --> 00:03:39.212 So the failure for regulated business is that not not that the mistake happened but nothing stood between the agent and its judgment and the API. s37 00:03:39.212 --> 00:03:42.492 So there were no real guardrails in place. s38 00:03:42.572 --> 00:03:51.932 Before we get into kind of the blueprint of how we can do better here, I wanted to give you three numbers that represent how this is playing out in our industry. s39 00:03:51.932 --> 00:03:59.052 The first 82% of all organizations already run AI agents, no surprise probably to anyone. s40 00:03:59.052 --> 00:04:02.172 32% had an AI governance policy in place. s41 00:04:02.172 --> 00:04:05.974 So probably more surprising given the difference between the two. s42 00:04:05.974 --> 00:04:15.534 Um and then finally, when a breach does happen, thanks to Shadow AI of some kind, uh the average cost is usually calculated to be about $5 million or so. s43 00:04:15.534 --> 00:04:17.414 So, that that that's a fun stat. s44 00:04:17.414 --> 00:04:26.614 Um but essentially, we can conclude that um air adoption is universal, but the control layer doesn't really exist and it's definitely not universal. s45 00:04:28.254 --> 00:04:39.734 Um before we go to looking at what a control layer for agents could be in the scenario, um what let's actually just start at what we've asked people our employees to do s46 00:04:39.734 --> 00:04:41.294 uh for as long as we've known. s47 00:04:41.294 --> 00:04:47.294 We've held people to this kind of standard as long as there's been software or you know, SaaS software of any kind. s48 00:04:47.294 --> 00:04:51.854 Um the the things that we hold people to are you have to sign in as yourself. s49 00:04:51.854 --> 00:05:01.734 Uh we ask nobody we ask everyone to you know, not share passwords or keys of any kind uh and you can't approve your own escalation or changing permissions. s50 00:05:01.734 --> 00:05:04.174 And we expect systems to keep track of all changes. s51 00:05:04.174 --> 00:05:05.214 That's pretty standard, right? s52 00:05:05.214 --> 00:05:06.614 That's all software today. s53 00:05:06.614 --> 00:05:08.214 That's what we expect. s54 00:05:08.214 --> 00:05:19.454 Um the question uh if you're being audited in a regulated industry, the question that will be asked is, you know, not is the model smart enough is your agent smart enough or using the latest release, s55 00:05:19.454 --> 00:05:25.454 but it'll just be who acted, who allowed it, and then show me the record of what happened. s56 00:05:26.774 --> 00:05:32.494 So, if we're going to apply this to agents, the first thing I'll say is we don't want to stop agents, right? s57 00:05:32.494 --> 00:05:37.574 We don't want to uh create a block that teams are going to find a way to route around. s58 00:05:37.574 --> 00:05:38.694 Agents will work around. s59 00:05:38.694 --> 00:05:41.054 We know that blocks don't really work. s60 00:05:41.054 --> 00:05:47.494 Uh the answer is creating a path to action that is meaningful and uh much easier than the ungoverned one. s61 00:05:47.494 --> 00:05:51.334 So, we want the governed action to actually be easier than the ungoverned action. s62 00:05:51.334 --> 00:05:58.694 Um let me show you with another fun animation what that path end to end could look like for a truly governed action. s63 00:06:01.974 --> 00:06:11.295 So, imagine you had a true control layer, you know, a magical system of some kind that governed agents on your machine and in the cloud. s64 00:06:11.295 --> 00:06:16.135 Um the first thing that you really care about is that it works just out of the box. s65 00:06:16.135 --> 00:06:21.175 So, you join the company, you onboard, and you want all of these policies in place. s66 00:06:21.175 --> 00:06:31.655 Um so, in this kind of first step of the animation, we're we're showing that it's something that comes out of the box installed on your laptop the first day you start at this company. s67 00:06:31.655 --> 00:06:36.775 Um the policy of what should happen or not happen is written once centrally. s68 00:06:36.775 --> 00:06:41.495 Uh it's not something a developer needs to write on their agent or their own harness. s69 00:06:41.495 --> 00:06:46.095 Uh it's written once and now kind of enforced at the at the company level. s70 00:06:46.095 --> 00:06:48.295 Uh for this one, let's think of a simple one. s71 00:06:48.295 --> 00:06:52.095 We'll go with secrets on user prompts and tool requests. s72 00:06:52.095 --> 00:06:56.095 And the action that we want is block or a quarantine. s73 00:06:56.455 --> 00:07:06.015 Um we'll take the same scenario again um of kind of what we did earlier, uh you know, from the developer accidentally deleting the database. s74 00:07:06.015 --> 00:07:10.335 Uh in this case, um the user's putting in AWS key in. s75 00:07:10.335 --> 00:07:13.975 Um or it might even be coming from another agent. s76 00:07:13.975 --> 00:07:19.895 Before the model ever sees it, a hook fires, and um this happens all in the harness runtime. s77 00:07:19.895 --> 00:07:28.055 So, it detects that there's a key that's being sent, um and the prompt is caught by a policy that's evaluated by the control layer. s78 00:07:28.055 --> 00:07:32.735 Uh the policy is enforced by a detector, and one of them matches the AWS key. s79 00:07:32.735 --> 00:07:41.655 So, there's there is a concept of like an interception that happens, um and then as that happens, there's also a decision that's created. s80 00:07:41.655 --> 00:07:46.855 And all of that happens in, you know, ideally we want it to be like 100 milliseconds or less. s81 00:07:46.855 --> 00:07:52.215 We want We don't want the agent or the human to really change the the path that they're on. s82 00:07:52.215 --> 00:07:54.655 Um so, it happens very quickly. s83 00:07:54.655 --> 00:08:03.735 And then finally, um you know, once the interception happens, that's logged, and now there's an audit of that event that's happened. s84 00:08:04.055 --> 00:08:14.017 So, inspected, blocked, and logged, the secret never really left the machine, and in this case the developer lost nothing but just a bad prompt. s85 00:08:14.617 --> 00:08:18.777 Um let's simplify this kind of conceptually into what's happened. s86 00:08:18.777 --> 00:08:26.177 Um we will look at this as four steps that a company can model a governance architecture around. s87 00:08:26.177 --> 00:08:28.177 Uh the first I like to call connect. s88 00:08:28.177 --> 00:08:31.560 Um this is creating a one creating a door instead of, s89 00:08:31.560 --> 00:08:31.977 [snorts] s90 00:08:31.977 --> 00:08:34.697 you know, having dozens of different integration points. s91 00:08:34.697 --> 00:08:40.737 All of the calls that an agent does when it connects to external systems should go to one place, something like an MCB gateway. s92 00:08:40.737 --> 00:08:43.977 A lot of us have used those and understand those well. s93 00:08:43.977 --> 00:08:50.417 Um the second is that the calls need to be resolved to a named agent or a human behind it. s94 00:08:50.417 --> 00:08:54.377 Um policy evaluation doesn't really work without identity. s95 00:08:54.377 --> 00:09:01.217 Um this has to be the same identity all of the enterprise software is assigned to as well, so it's consistent across your organization. s96 00:09:01.217 --> 00:09:09.777 Um Third, there's a decide step where you either allow, deny, quarantine, or you escalate the decision. s97 00:09:09.777 --> 00:09:13.097 Um and then sometimes even just warning a user is sufficient. s98 00:09:13.097 --> 00:09:20.097 And then finally that decision is recorded um before, you know, any action is actually taken. s99 00:09:20.417 --> 00:09:28.737 Um we're going to dive into couple of these in a little bit more detail so I can give you a better blueprint of how to do this uh for your own product or your own company. s100 00:09:28.737 --> 00:09:31.137 The first is uh identity. s101 00:09:31.137 --> 00:09:35.937 So, who was on the call, who actually executed uh this request? s102 00:09:35.937 --> 00:09:41.657 Um if we look at this uh kind of screen in front of us, there are two sides to it. s103 00:09:41.657 --> 00:09:45.137 The left side is the incident that we posed. s104 00:09:45.137 --> 00:09:46.857 That's the log from the incident. s105 00:09:46.857 --> 00:09:51.377 A key was found on the file, but no, there's no actor and there's nobody who approved it. s106 00:09:51.377 --> 00:09:58.537 Uh so, if you're looking at this as, you know, uh an auditor, you're wondering who actually took the action, who's responsible. s107 00:09:58.537 --> 00:10:03.177 Um these And to be honest, this is the default state of most systems today. s108 00:10:03.177 --> 00:10:06.617 Um it was even ours you know, until earlier this year. s109 00:10:06.617 --> 00:10:12.458 Every principal, every agent, every user, um it's all borrowed based on a person's identity. s110 00:10:12.458 --> 00:10:16.138 This is pretty common in most companies today running agents. s111 00:10:16.138 --> 00:10:23.658 Even a background job is mapped to a user or there's just a anonymous kind of system role that's created. s112 00:10:23.658 --> 00:10:27.778 Even Slack assistants or tend to be running as like user permissions. s113 00:10:27.778 --> 00:10:30.338 They don't necessarily have their own identity. s114 00:10:30.338 --> 00:10:34.578 On the right is that same prompt now modeled to the lens of a control layer. s115 00:10:34.578 --> 00:10:38.538 The agent is a principal in its own right, so it has its own identity. s116 00:10:38.538 --> 00:10:44.778 Below that you see there is an an operator who drove the action. s117 00:10:44.778 --> 00:10:47.778 So who is actually interacting with that with that agent in real time. s118 00:10:47.778 --> 00:10:49.258 In this case it's me. s119 00:10:49.258 --> 00:10:52.538 Um and that's recorded for attribution. s120 00:10:52.538 --> 00:10:55.778 It's never used to like widen access in real time. s121 00:10:55.778 --> 00:10:57.618 And then finally there's the authorizer. s122 00:10:57.618 --> 00:11:01.978 So this is the human who approved the original credential setup for the agent. s123 00:11:01.978 --> 00:11:04.298 In this case it says Thomas who's our CISO. s124 00:11:04.298 --> 00:11:06.778 This is fixed for credentials lifetime. s125 00:11:06.778 --> 00:11:09.418 So the right hand card is what we really want for compliance. s126 00:11:09.418 --> 00:11:15.538 This is a true log of what actually happened and how we could model it. s127 00:11:16.298 --> 00:11:25.458 Um the next one is for kind of the set theory nerds in the audience, but it's how you actually decide in real time whether the call succeeds or fails. s128 00:11:25.458 --> 00:11:28.538 Um let's the kind of three parts to this. s129 00:11:28.538 --> 00:11:31.738 The first is the one on the left. s130 00:11:31.738 --> 00:11:33.098 Our credentials. s131 00:11:33.098 --> 00:11:36.258 This is the ceiling of what a human put down when they issued the key. s132 00:11:36.258 --> 00:11:39.138 So this again is the thing that never gets widened. s133 00:11:39.138 --> 00:11:43.578 It lives in the credential and it's something you can't escalate past. s134 00:11:43.578 --> 00:11:48.018 The second is we call it A. It's the agent's own policy. s135 00:11:48.018 --> 00:11:53.218 It can't hold an admin privilege or any or any kind of human only scope. s136 00:11:53.218 --> 00:11:55.458 And then finally O is the owner's live policy. s137 00:11:55.458 --> 00:12:01.458 So an agent can never do more than the person accountable for it who who's actually driving the agent. s138 00:12:01.458 --> 00:12:03.138 So in this case it was me. s139 00:12:03.138 --> 00:12:10.738 Um what's interesting about modeling it this way is that you know, the second and third A and O are read on every call. s140 00:12:10.738 --> 00:12:13.699 If you remove a permission, um, it's gone on the next request. s141 00:12:13.699 --> 00:12:15.459 So, it's re-evaluated. s142 00:12:15.459 --> 00:12:24.379 Um, and if we go back to that database scenario that we started with, even if the owner could delete production, an agent scoped to just staging can't. s143 00:12:24.379 --> 00:12:29.899 And if someone gave the agent a broader policy, it can't exceed an owner who never had production rights. s144 00:12:29.899 --> 00:12:32.579 So, in kind of both directions, you're protected. s145 00:12:32.579 --> 00:12:34.739 Um, either way the call would stop. s146 00:12:34.739 --> 00:12:48.179 Uh, and then finally, an important note, you know, if an if an owner would leave a company, um, since it's connected with your enterprise identity, your IDP de-provisions them, and every request that agents make is now denied going forward. s147 00:12:51.619 --> 00:12:56.259 Uh, the last step in that conceptual flow is is recording what actually happened. s148 00:12:56.259 --> 00:12:59.539 Um, so, on the left is again that scenario playing out. s149 00:12:59.539 --> 00:13:05.379 Um, on the right is what it actually leaves behind, now that we have a way to kind of model the steps that have happened. s150 00:13:05.379 --> 00:13:14.019 Uh, we can see kind of in this flow, as you go through the different steps, there's a denial step that happens, uh, and there's a stated intent, which was that, s151 00:13:14.019 --> 00:13:18.299 uh, the agent wanted to act on staging, but the action was done on production. s152 00:13:18.299 --> 00:13:20.739 Um, the policy sees both sides of that. s153 00:13:20.739 --> 00:13:24.019 So, this is where compliance frameworks have generally landed today. s154 00:13:24.019 --> 00:13:33.379 Um, every non-read action is evaluated against policy uh, that considers the agent's stated intent, uh, not just the action that actually happened. s155 00:13:33.379 --> 00:13:44.859 Um, in most control systems, there's exactly one of kind of five decisions that comes out of this, uh, like we mentioned earlier, it's either allowing, denying, modifying, or escalating up to a human. s156 00:13:44.859 --> 00:13:56.259 Um, every evaluated action produces this kind of record of what's happened, and then in generally, in a lot of systems, it's cryptographically bound to the agent identity. s157 00:13:56.859 --> 00:14:02.379 So, I have only one slide that's going to actually take you through how we built this at Speakeasy. s158 00:14:02.379 --> 00:14:04.379 So, I'll keep it limited to that. s159 00:14:04.379 --> 00:14:08.899 Um, as I mentioned earlier, we we Speakeasy were an AI control plane. s160 00:14:08.899 --> 00:14:17.101 Uh, we actually put our product in between agents uh, that people use or using autonomously, and the systems that they have access to. s161 00:14:17.101 --> 00:14:21.701 Um so the identity comes the identity baseline comes from your IDP. s162 00:14:21.701 --> 00:14:24.101 Uh so Okta or another identity provider. s163 00:14:24.101 --> 00:14:28.941 Uh and then the policies of action is actually evaluated on our product or our control plane. s164 00:14:28.941 --> 00:14:34.701 So every call um includes the ones including the ones that we deny all go through us. s165 00:14:34.701 --> 00:14:37.821 And so we write out the audit log for for our customers. s166 00:14:37.821 --> 00:14:41.181 There's kind of three kind of critical decisions in this architecture. s167 00:14:41.181 --> 00:14:45.261 Uh the first as you can see is that we sit in the critical path not beside it. s168 00:14:45.261 --> 00:14:48.141 Uh the record has to be of what actually happened. s169 00:14:48.141 --> 00:14:57.101 Uh not what the agent reported and two we use kind of first class identity uh instead of using user logins everywhere to the system. s170 00:14:57.101 --> 00:15:00.021 And then finally we'd actually distribute to MDM. s171 00:15:00.021 --> 00:15:02.421 Uh so all of this comes working out of the box. s172 00:15:02.421 --> 00:15:05.901 For a user they don't have to go redefine these policies on their end. s173 00:15:05.901 --> 00:15:11.861 It's something that just works the moment they start uh at the company or start using an agent. s174 00:15:11.861 --> 00:15:18.061 So that's my only slide about Speakeasy but I'm happy to talk to you more about that later. s175 00:15:18.301 --> 00:15:23.061 Um So that's the to kind of summarize what we went through. s176 00:15:23.061 --> 00:15:32.061 Uh in a regulated industry permission to act usually means kind of three things have to be shown for every agent or every agent action that is taken. s177 00:15:32.061 --> 00:15:37.621 Um who the actor was, what policies allowed it uh and the signing of the record. s178 00:15:37.621 --> 00:15:40.621 So knowing for a fact that this actually happened. s179 00:15:40.621 --> 00:15:46.741 Um this is what we've always kind of asked employees to do at a business and it's what we ask agents to do now. s180 00:15:46.741 --> 00:15:49.221 Um valid credentials are not valid intent. s181 00:15:49.221 --> 00:15:55.741 That's something that's really important to remember and the fix is not you know just saying we have a smarter model or a better agent. s182 00:15:55.741 --> 00:16:01.941 Uh it's actually putting together identity policy and record of every action that's that's happening. s183 00:16:01.941 --> 00:16:05.421 So that's uh kind of everything I wanted to talk to you about today. s184 00:16:05.421 --> 00:16:18.141 Um I'll also leave you with um we're doing an event tonight with our friends here at Tailscale um launching a partnership around private put taking your MCP gateway and putting it on a private network. s185 00:16:18.141 --> 00:16:20.182 Join us if if you're here. s186 00:16:20.182 --> 00:16:22.902 And yeah, happy to take questions. s187 00:16:23.821 --> 00:16:25.821 [applause] s188 00:16:37.022 --> 00:16:37.942 Go ahead, camera. s189 00:16:37.942 --> 00:16:50.222 I had one question on say you're using an agent that is being governed by the ethics of the planet we're talking about and disagree with the decision that it came to to deny that that action. s190 00:16:50.222 --> 00:16:52.902 What would you How would you design a policy? s191 00:16:52.902 --> 00:16:54.942 Yeah, that's a that's a good question. s192 00:16:54.942 --> 00:17:03.862 I think um the you know, if probably the first most important thing is that you do have a way to see what decisions were taken. s193 00:17:03.862 --> 00:17:15.942 So, in most companies today when people are using you know, just let's just say Claude from Anthropic and connected to all the systems, as it's accessing these systems, there's really no record of those decisions taken. s194 00:17:15.942 --> 00:17:26.062 And so, even if you disagree that hey, this user shouldn't have been able to access, let's say your CRM, you can't go back and actually say that has happened for sure. s195 00:17:26.062 --> 00:17:34.622 Secondly, you do want a way to escalate your user when that happens so you can actually decide if you want that to be passed or not. s196 00:17:34.622 --> 00:17:38.422 So, those I would say they're kind of more fundamental missing pieces. s197 00:17:38.422 --> 00:17:43.942 If you have that in place, you can always go update a policy to actually allow that to happen. s198 00:17:46.582 --> 00:17:47.782 Yeah, go ahead. s199 00:17:47.782 --> 00:17:57.062 Out of the 56% without policies, I'm curious who's buying because regulators force them versus something broke. s200 00:17:57.062 --> 00:17:59.182 Like, what moves the budget? s201 00:17:59.182 --> 00:18:00.102 Yeah, that's a good question. s202 00:18:00.102 --> 00:18:02.582 Something we think about a lot. s203 00:18:02.582 --> 00:18:15.742 I would say compliance is a really good driver for people to think about getting a solution in place, but I think the the reason what I want to leave you with is that it you don't necessarily s204 00:18:15.742 --> 00:18:24.343 you know, buying a product is to fulfill compliance, but the first thing you need to do is kind of understand what is actually happening in your business first. s205 00:18:24.343 --> 00:18:27.743 So, I would say compliance is a great driver of budget. s206 00:18:27.743 --> 00:18:33.543 When there is an incident that that often, you know, raises the priority of something like this. s207 00:18:34.583 --> 00:18:35.583 Go ahead. s208 00:18:35.583 --> 00:18:40.703 Do you mask PII or like the logs that come through your control plane? s209 00:18:40.703 --> 00:18:41.183 We do. s210 00:18:41.183 --> 00:18:47.023 We do everything from kind of blocking, masking to redacting PII. s211 00:18:47.023 --> 00:18:55.023 Um, what we've found in most businesses when you turn on observability and you look at everything going through the agents, there's a ton of PII going through. s212 00:18:55.023 --> 00:19:00.863 So, if you are going to kind of block on PII, you'll probably grind your business to a halt, right? s213 00:19:00.863 --> 00:19:04.983 It's just it's hard for people to not use the agents with PII. s214 00:19:04.983 --> 00:19:09.223 I think the more important thing to look at is what is happening to that PII. s215 00:19:09.223 --> 00:19:10.863 So, like what's the intent for the agent? s216 00:19:10.863 --> 00:19:17.743 Is it going to post it in a public Slack channel, send it by email, put it back into a tool called to another system? s217 00:19:17.743 --> 00:19:29.383 Those are the kind of policies we try to recommend people to, you know, use observability to define rather than just kind of blocking or redacting all PII together altogether. s218 00:19:29.383 --> 00:19:37.503 I guess I'm asking that question because it's like if there is an incident, I'd ask the control plane for like our logs and to understand what happened, right? s219 00:19:37.503 --> 00:19:38.663 With authorization? s220 00:19:38.663 --> 00:19:40.263 Yes. s221 00:19:40.263 --> 00:19:42.583 You would have to like give those logs to me, right? s222 00:19:42.583 --> 00:19:46.873 But, why should I give you like the data s223 00:19:46.873 --> 00:19:47.103 [snorts] s224 00:19:47.103 --> 00:19:47.783 in the first place? s225 00:19:47.783 --> 00:19:50.423 Like because it's proprietary and it's our secret sauce. s226 00:19:50.423 --> 00:19:50.863 Yeah. s227 00:19:50.863 --> 00:19:53.423 So, like what's the trust model with that? s228 00:19:53.423 --> 00:20:02.743 Yeah, I mean, that's kind of it's getting a little into the details of how we deploy, but like we do deploy our software in a few different ways. s229 00:20:02.743 --> 00:20:08.823 Like we do offer kind of a data plane option where it runs in your own VPC, so data stays with you. s230 00:20:08.823 --> 00:20:13.223 But, the more common option that customers do is they bring their own encryption keys, right? s231 00:20:13.223 --> 00:20:20.543 So, we do encrypt the data, but you bring your own keys for so you can govern it yourself. s232 00:20:20.543 --> 00:20:30.265 And yeah, customers often will go into kind of our product and look at what is all the PII at that happened at least look at from the perspective of like all the findings s233 00:20:30.265 --> 00:20:34.345 even if they're not looking at the redacted data itself.