Name: Product Launch: Agentic AI & Data Security Expansion
Uploaded: 2026-05-05T18:49:11.773Z
Duration: 31 min 6 s
Description: Product Launch: Agentic AI & Data Security Expansion

Transcript for "Product Launch: Agentic AI & Data Security Expansion": Something fundamental changed in the last twelve months. AI has crossed the threshold from a tool you prompt to an actor operating inside your environment. Today, AI agents read email, ship code, and query databases. And increasingly, they act without human oversight. Sensitive. Cyberhaven Labs found that nearly 40% of AI interactions involve sensitive data. And the average employee shares sensitive data with AI tools once every three days. What makes this moment different from every prior wave of disruption is this. In the network error, data had a home, and security had a boundary. In the cloud error, that boundary dissolved. In the agentic error, the human is no longer guaranteed to be in the loop. Agents don't wait for approval. They take actions and chain workflows across systems faster than security teams can track. Prompt injection, tool poisoning, workflow hijacking, these aren't theoretical risks. They're happening right now. Data doesn't move neatly in defined categories or labeled files. It moves across endpoints, cloud apps, and now AI agents, fragmented and constantly changing. Securing it requires understanding its full journey, where it lives, how it moves, and what's happening along the way. This is what CyberHaven was built to do. Today's release extends the cyber haven AI and data security platform across two fronts. The first is the AI security front. AI agents are already operating across your enterprise in developer tools, productivity suites, and custom workflows. The challenge is that most organizations have limited visibility into what their employees are using, limited monitoring of what agents are touching, and very limited controls to intervene before something goes wrong. We are addressing that directly with agentic AI security built on our data lineage foundation and the analyst plug in that puts cyberhaven's intelligence directly into the hands of people investigating threats. The second front is expanding the data security foundation itself. AI adoption has changed which devices matter to security teams. Contractor machines, Chrome OS environment, unmanaged devices, these are now all part of the attack surface. The standalone browser extension extends Cyberhaven coverage to all of them. And expanded screen capture gives insider risk investigators something they have never had before, the full picture, not just the moment of an incident. We will start with the AI security and what it actually takes to govern agents in production. When we talk to CISOs about agentic AI risk, a few themes surface immediately. Frequently, they don't know what agents are running. They may know about Cloud Code because IT approved it, but they don't know about the local OpenClaw installs or a rogue or a custom MCP server a team has spun up to automate a deployment pipeline. These are the shadow agents and shadow MCP servers. They often run on endpoints, not in the cloud, which makes their initial deployment completely invisible. Even if teams know an AI agent exists, they can't see what it's doing at the level that matters. An agent manipulated through prompt injection doesn't reveal intent in any single message. It reveals itself across an entire conversation thread across a chain of tool calls and that blind spot grows when multiple agents are working together. An AI agent running on an employee laptop inherits the employee's identity, gains access to sensitive cloud data, and can bulk read records from trusted sources such as production database and move sensitive data to various public destinations. Many controls today are blunt. Block first policies generate alert fatigue, drive employees towards personal accounts, and create friction without meaningfully reducing the risk. Cyberhaven's agentic AI security solution addresses all with three integrated layers. Visibility, a continuous automatically maintained inventory of every agent, Genya app, MCP and AI connectors across your environment on endpoints, in IDEs, in SaaS and across all MCP connected systems. Every app and agent is assigned a risk IQ score across five dimensions. And also, personal versus corporate usage are detected at run time. The next one is observability. We reconstruct the full execution life cycle of every agent interaction, the data it has accessed, the tool calls it's invoked and the actions it took, and the complete multi turn conversation context more of a flight recorder. Violations emerge across multiple agent steps, sometimes across agents working in sequence. Cyberhaven can correlate the full conversation and execution thread out of the box, giving security teams forensic context without manually reviewing every conversation. And the last but not least, the controls. The runtime policy guardrails that stop high risk actions before they cause damage. Guardrails can be configured to block, warn, or redact at the prompt and response level. We replace generic block pages with plain English explanation of what the risk is and why. We warn users when a prompt contains sensitive data with the option to revise or proceed. The goal is governance that doesn't kill productivity. Instead of blanket blocking, users receive contextual guidance that explains the risk and gives them a path forward. The differentiator that ties visibility, observability, and controls together is the data lineage. Every other agentic AI tool tells you what an agent did. Cyberhuman tells you where the data came from, what it contained, and where it went next. That's the difference between an alert and an investigation. And it's it's only possible because this solution is built on a lineage foundation that has already been mapping your data for years. Ask any security analyst what their morning looks like, and the answer is usually the same. Before any real investigation begins, there's an hour of assembly work. Prioritizing which incidents actually matter, validating true positives using lineage and classification, tracing activity across users, systems, and data to understand scope. Each step means jumping between tools and making calls with incomplete context. Every step is a context switch. Cyberhaven recognized this early and has been a leader in applying AI agents to solve it with Linear AI, an always on analyst that detects unknown risk, prioritizes incidents and accelerates investigations. Linear AI, things like the smartest security analysts, understanding how people across your organization, and identifying risk even without predefined policies. The analyst plugin extends the same agent driven approach into the tools your analysts already use. It connects cyber haven signals including data lineage classification and incident findings directly into the AI tools your analysts already have open. Cloud Code, codecs, or an MCP compatible client, your AI assistant becomes a security analyst with full access to your Cyberhaven data. Cyberhaven analyst plugin ships with 40 plus pre built security skills that run complete multi step workflows and 20 plus specialized analysis agents that serve as the engine underneath each one. Analysts don't need to memorize them. Describe what you need and the plugin routes to the right workflow automatically. The triage agent works through incidents. The compliance agent maps finding to regulations. The executive agent translates risk. Others handle posture scoring, anomaly detection, Gen AI exposure analysis, and false positive root cause identification. All of them run automatically with the native understanding of cyber havens data model, lineage and incident structure. Skills don't just surface findings. They can close duplicate incidents, execute actions, and preview policy changes before they go live. Let me walk through how this works in practice. You start the day and ask, review today's incidents on production. Instead of manually pulling a queue and cross referencing context, you get a prioritized report. What happened, what data was involved and recommended next steps. Mid investigation, show me everything connected to the source code leak flagged this morning. The plugin traces outward from a single incident to related users, infected host, and destination systems, giving the analyst full scope of exposure. End of the month, generate an executive briefing for the last month. Incident trends and compliance posture are compiled automatically. These are live workflows running on your cyber haven environment. For teams that want to connect their own tooling, the remote MCP connector handles authentication and data access centrally. The next two capabilities focus on the data security foundation, ensuring the platform reaches every device where sensitive data is at risk. Until now, Cyberhaven browser extension worked in tandem with the endpoint sensor. The standalone browser extension changes that, extending coverage to any browser capable device without requiring an endpoint deployment. It traces uploads, downloads, and copy paste in the browser with content inspection applied. It captures cloud account context to distinguish corporate from personal usage. And it is managed from the same sensor management interface as endpoint deployments. Chrome OS is now a primary platform for entire organizations, especially in education, health care, and distributed enterprise teams. The standalone browser extension brings Cyberhaven coverage to all of them with the same content inspection, cloud account visibility, and policy enforcement available on any other managed device. Contractor and unmanaged devices are similar story. They are the devices that access corporate systems, handle sensitive personal data, and operate outside traditional security boundaries. The standalone browser extension protects them without requiring IT to manage an endpoint deployment on devices the organization does not own. Teams can extend coverage to browser based movement across unmanaged devices and start seeing real activity within hours. Expanded screen capture takes our existing screen capability significantly further, and it matters most to the teams running insider risk investigations. When a security team is investigating a high risk event, context is everything. What was the employee doing in the minutes before the alert fire? What happened immediately after? Video based capture now covers that full window around the event, giving investigators a continuous record rather than isolated frames. Security teams can also designate specific users or devices for continuous capture independent of any incident trigger. For users already on the radar, evidence collection begins before a formal investigation opens. Together, these capabilities give insider risk teams faster, richer evidence at every stage of an investigation. Security isn't about just where data lives anymore. It's about how it moves across systems, across users, across AI agents. And the only way to secure it is to follow it. To understand where it came from, how it changed, and where it goes next, that's what Cyberhaven makes possible. Agintiq AI security, analyst plug in, stand alone browser extension, and expanded screen capture. Each one advances Cyberhaven AI and data security platform to where data is moving today. Together, they give security teams the foundation to govern data wherever it lives and goes in this agentic world. Hey, everyone. It's, Bruce Chen and Kasi Annamalai live. We're getting some questions in chat. We're getting some questions in the q and a tab. Help us out by dropping the questions in the q and a tab so we can better sort and organize, which questions are coming through. Kasi Annamalai, there's a one. Do you wanna take that one? Yeah. Absolutely. The question is, just reading it loud, for folks, how is Cyberhaven AI security different from other tools? Great question. Thanks for asking. I mean, when it comes to AI that employees are using, organizations don't know what they don't know. Right? So that's a blind spot. And there's a lot of solutions probably existing ones you have. We'll discover the Gen AI tools, Gen AI usage that runs on the browser, but that's about it. Right? Especially in the new agentic world, where Cyberhaven provides the GenAI discovery and visibility already for more than a year now. And now what we are expanding is getting the visibility into the agents that are hidden, specifically the locally installed ones. And now you will see the agent activity related to that as part of the agent in AI security that we announced now. Beyond the browser, beyond, the network that traditional tools and many companies have at their disposal, but what you're missing is the agents installed on the laptops. Right? So that's what we are discovering now and presenting it, which we believe, with the context and the data that Cyberhaven already have, is is more, helpful for customers to understand these shadow agents that employees are using in their laptops. So that's one of the Cyberhaven's latest AI solution, that is addressing both of these gaps. Thanks, Kasi. There was another one in the chat that I wrote down earlier. So someone asked, is there a Cyberhaven MCP available yet? If so, where can we access it? I think they asked it when we were presenting the, the analyst plug in. So that's what that is. So the plug in ships, you know, today with a local MCP server, and, we'll be deploying that MCP server along customer's tenant, in the June release. Alright. Let's just, cause you guys scroll through the thing and answer questions here. So So let me just Absolutely. scroll. One other question in the chat, which endpoint operating system, are available? We support most of the, major ones, Windows, Mac, Linux. Again, if there is anything specific, please let us know. Happy to answer that. That's another question that we had on the chat. Bruce, there was one question, I think, from the earlier one related to the MCP server. Do you wanna take that? I may have missed that one. Let's see. I can read it out. The question was, is there a MCP server for Cyberhaven available yet? If so, where can we access, yeah. it? Yeah. Yeah. That that one I answered first. Yeah. I think yep. So there there's another one about how that's different from just using Claude that one's in the q and a box. So, I can answer that. So if if you have an analyst that's, Okay. already just using Claude, like, you know, Cowork or they're using ChatGPD, you know, I I think what's different from, I'm talking about the analyst plug in from using, let's say, just your standard, cloud. Yeah. I I think what's fundamentally different is just gonna be the context. Right? So if you're just using chat gbt, you're using cloud, you're asking a question, you know, it doesn't have all the telemetry that, cyber cyber haven has. Right? But if you use the plug in, you get all that telemetry. You get the lineage. You get the data classification. You get the, events. You get the information on the users if they're high risk users. Right? So so that's really the big difference there. You know, you can use the analyst plug in and you can say, hey, you know, what's the boss radius of this incident or, you know, who's, exfiltrating, IP data this month. Right? And and so that's that's all the the questions. That's all the kind of, intel that you can get from the plug in that you wouldn't get from just using, you know, an AI assistant that's really not plugged into your security environment. Alright. Let me take a look at the other q and a stuff here. Alright. I see there's some conversations, about the stand alone, browser extension. Let's see. I'm just reading I think is it Tim's comment and then Prashant's comment here. So let me just elaborate a little bit about that. I I think, you know, one of the use cases that when we're building this product and we're talking to customers about is, the contractor use case. Right? It's all these unmanaged devices. It's it's, you know, an enterprise, you might have a couple 100, couple thousand contractors who, you know, they don't want an endpoint agent installed on, their laptops, and it's just not practical. Right? And so by having this, standalone browser extension, which we've always had, it's just you know, we've updated it now. So now you we call it the standalone browser extension. So you no longer need the agent. All you need is the browser extension, and you can start getting visibility on what's happening on those contractor devices. Right? You can see uploads, downloads, that we talked about earlier. You can see, further telemetry like, you know, who that user is, the URL that they access. And so you're getting a lot more visibility, now with just having the browser extension and you do not need an endpoint sensor, anymore, at least for those use cases. Okay. Kasi. There is one more question about, the MCP server, detection at the endpoint, and how how can you make sure the MCP servers can be, detected, especially when they can be renamed and no standards. Great question. I can quickly, cover that. You're absolutely right. The approach that we are taking is to, get closer to the deployment of the agents at the employee laptops and be able to discover and detect all the MCP tool calls that's being made, whether it's local MCP or remote MCP, that is in use from the agent perspective. So that's the reason even if there anything changes, on the MCP server that's getting deployed, we will be able to detect that as long as it's in use. So we are monitoring the agent activity and any tool calls that's being missed made. So we are sitting closer to the application. We are connecting to the application, hook. So we would still be able to extract those telemetry, and that's the reason, like, some of the man in the middle, proxy solutions, may may have, I mean, difficulty passing through that. But we, sitting closer to the endpoint, and the application, we are able to detect that. Great question. Thank you. There was one more question which I answered live, but, I mean, in the chat, but can quickly cover here as well. Do you also perform AI SPM, AI security posture management, for, I mean, for apps like Cloud Cloud Code? Short answer is absolutely yes. We will be able to get the partial related info extracted, what plugins are in use, what skills are in use, what MCP server is in use, and also all the risk assessment related to that. Is it a legit MCP server, or is it a skill that is published, at the organization level and all of that information? So we would be able to present you with posture related, visibility and the risk assessment, related to that. Alright. So the last feature, you know, we talked about four features here. The last feature we talked about was the expanded screen capture. I just wanna clear up any confusion that might be there. So, you know, we've had the screen capture capability now for a while. And, you know, the scenario here that you know, for why we built this was, you know, if you imagine an employee who's planning on leaving the organization. Right? And they might not have, you know, told you yet that they're leaving the organization, but maybe they put in there two weeks notice at some point in time. Well, you can actually go back in time, and you can see a window of activity, recorded window of activity, you know, when they may have, exfiltrated, data, when they may have, like, sent, information, you know, to their personal, accounts. So so you just have this, you know, this longer window of time of seeing potential exfiltration activity. You know, we really built this for, investigating, insider risk, use cases. So that is what the expand expanded screen capture, is. Again, it's already, you know, we've had variations of this feature before, but this just allows you to have a video based, recording now if you choose to use that feature. Okay. There is one more question. Does the agent monitoring capability include monitoring, logging, reasoning, traces where available? Not the reasoning, but everything at rest, that you mentioned, the monitoring, logging, the tool calls, the agent activity, the internal commands, if the agent especially if it's a local agent, trying to make a file modification in the local machine, we would be able to capture that, and be able to present it in the, conversation visibility, as part of the agent accession. So we do capture all of those agent activity, files that has accessed, MCB tool calls that's made, any internal commands that has been executed, all of it. Maybe, Kasi, if you have time, if you see any other questions about the AgenTek AI that you wanna take, and then maybe I can answer one more question about the analyst plug in. No. Please go for it, Bruce. I'm scrolling over the question just to make sure I haven't missed any. I'm checking it. But if you have, please go ahead on the analyst plug in. Mhmm. Okay. Yeah. So about usability for the analyst plug in, how do you use it? There's really no learning curve to that. You know, that's the beauty of the analyst plug in. It's if you've already used ChatGPT, then it's then you already know how to use the analyst plug in. Right? We actually have onboarding, prompts in there to ask, you know, for example, what your role is if you're a CISO versus, like, a l one analyst. And then from there, it'll give you kind of a couple options on what you might wanna accomplish. Right? Do you wanna investigate a specific user? Perhaps if you're an analyst, if you're a CISO, it'll ask, hey. Do you wanna generate an executive report on incidents in the past three days or past quarter? So it just makes it really easy for you to, get the information you need about, you know, about data security, incidents or or exfiltration incidents. Thanks, Alright. Bruce. There is one more question, real quick from Tim. What happens to the s three bucket of the Google Drive? Are you gonna store all of this to that, and will there be any cost implication? The short answer is, it's again option, available to the customers. If you want, Cyberhaven to capture the evidence as part of a agent take workflow or agent take conversation, we would be able to capture that and then store it to the whatever their destination that you want us to. But, again, that's optional. Customer would be able to choose it, if you want to capture the evidence, whenever there is a data exchange, in a agent accession. Yeah. Maybe I'll answer one more question. I see one from Matt. Hey, Matt. Good seeing your name again. You know, is the question is is the AI features a new license model or will these features fold into which SKUs? Well, so maybe the way to to answer this is if you're you know, we released AI security last year. It was able to cover, you know, hundreds of gen a n I gen AI apps. If you're already using that feature, this is, you know, you're able to also now see agents, as well. So that's all kind of baked in together. I I don't know if, Kasi, you have anything to add about that, but I I know time's up. So No. You summarized it, Bruce. Thank you. Okay. Awesome. I know there's still more questions coming in. We'll go ahead and follow-up with everyone. You know, if you have a account manager or your customer service rep, then we'll go ahead and, answer the questions through them. Thank you everyone, for your time.