Name: Smart AI Security. Full Data Control.
Uploaded: 2026-03-11T17:46:06.405Z
Duration: 30 min 12 s
Description: Smart AI Security. Full Data Control.

Transcript for "Smart AI Security. Full Data Control.": Cloud and AI are revolutionizing work. Companies are scrambling to safely adopt new AI solutions, to drive automation, and to enable work from anywhere and anything, all to stay competitive in today's fast changing world. And the pace of change is only accelerating so quickly that it is outpacing existing security and networking tools. Twenty years ago, handling security and networking was like managing a single road, slow and straightforward. Then came the cloud, and that single road became a highway, faster, busier, with onrains, junctions, and interchanges requiring much more careful control. Who is entering? Where are they going? Today, AI adds an entirely new dimension. Now it's like controlling global air traffic. AI and cloud aren't static. They are dynamic, data rich, and interactive. They leverage new languages, services, and protocols. They seamlessly interconnect. They're instantly on, spun up, streamed, intelligent, autonomous, and are producing nearly 75 times more data than fifteen years ago. This is like thousands upon thousands of high speed flights requiring precise coordination, real time awareness, and deep understanding. Legacy security and first gen tools just weren't designed for this. Controlling the crowded skies of today requires an entirely new approach. Netskope was built from the ground up to redefine security and networking for the cloud and AI era and whatever comes next. Hello, I'm Sanjay Barry, CEO and co founder at Netskope, a leader in modern security and networking for the cloud and AI era. In 2026, organizations are undergoing faster, more comprehensive digital change than they have ever experienced before. Moving at much greater speed than the internet and cloud transformations that came before it. AI is ripping up the rule books in many industries, requiring a complete rethink of business models, service delivery, and organizational value. Driving this AI revolution? Data, the context that makes that data useful, and the need for speed. And what puts the success of this revolution at risk? A lack of proper AI security and guardrails for sure. But there's also a risk that legacy security methodologies apply to brakes far too heavily on the turbo possibilities that AI promises. Netskope is here to make security the accelerator, not the obstacle. At Netskope, relentless innovation has always been our core DNA. I often say that we skate to where the puck is going. We don't design for the world as it is. We build for the world as it will be. When we founded this company, we anticipated a massive multidimensional shift. We saw that the boundaries of the office and applications would disappear, that work would happen from anywhere, cloud would be the norm, and that traffic would evolve far beyond human users, to include an explosion of machine to machine and non human interactions. We didn't just build a security platform, we built a global cloud native network, an architecture purposely designed to secure and accelerate every interaction, no matter who or what is behind it. We called it Netskope One. The AI revolution represents the ultimate realization of that vision. Our architecture was built specifically to power this high performance environment. AI makes the world more complex. It demands a level of context that legacy systems simply cannot provide. How do you distinguish between a trusted employee and an autonomous AI agent? How do you separate a legitimate innovation query from a sophisticated prompt injection attack? Without that deep context, organizations face a false choice: block innovation to stay safe, or embrace AI and watch their intellectual property and security posture erode. At Netskope, we reject that trade off. We enable you to say yes to the future because our Netskope One platform protects the company and its data while letting them leverage the power of AI. We provide real time context aware protection against generative AI driven threats and data leakage without adding unnecessary latency. Whether it's a remote employee using an agentic browser, or a user or AI agent hitting a LLM, we provide the visibility, control, and security needed to innovate at the speed of thought. Netskope's elimination of the AI security latency tax and acceleration of the AI experience is partially made possible because of New Edge. Our global New Edge network is powered by over 80 plus regions and 120 plus data centers, and all Netskope services run and are available to all customers in every data center. Our extensively peered and optimized network places your users and AI agents just milliseconds from the generative AI apps, LLMs, GPUs, and CPUs powering AI. This isn't a vision for tomorrow. It's the reality for over 1,000 of Netskope's customers using Netskope One to secure AI. They are already using our platform as their AI command center, gaining visibility and control into corporate managed and shadow AI, understanding generative AI app risks, as well as protecting against gen AI threats and sensitive data from being uploaded into generative AI or from being used to train LLMs inappropriately. Today, we take the next step in that journey. I'm thrilled to introduce our expanded AI security functionality, providing an advanced suite of tools built for the borderless AI driven world. It recognizes the multifaceted complexity of securing and accelerating AI, and captures our belief that in order to be optimized, security needs to be data centric, unified, and delivered through one of the most performant, resilient, and interconnected networks in the world today. Today, we are announcing new innovations and products as part of Netskope One AI Security, designed to solve the challenges of secure AI adoption. AI visibility, over what is being used and where data is going. AI control, over the access AI models are given. AI protection, against AI specific threats. AI guardrails, over how AI uses the data it is allowed to access and AI development, building secure by design AI models and agentic infrastructure. NETSCOPE ONE AI Security is designed to enable technology teams to say yes to AI innovation, starting with AI experimentation and building towards a true AI advantage. This is the AI Fastlane. Smart, safe use, total data control. To talk you through the new product components of our NETscope one AI security solution, and show how they integrate seamlessly into your existing Netskope One platform capabilities. I'm gonna hand you over to Jess Leroy and Melody Nuri from our product team. We've all felt it, that intense pressure to accelerate with AI. Security leaders, you aren't just managing risk anymore. You're reinventing business processes. You're taking AI exploration from the sandbox and moving it to the core of your service delivery. And month by month, we see the needs of your organization shifting as you move deeper into this AI maturity journey. As Sanjay mentioned, Netskope is here to be your guide. We spent the last decade mastering the human to application flow, and our secret sauce has always been granular context. Netskope doesn't just see and secure a user hitting a Gen AI app, website, or cloud application. We see a specific identity on a managed device performing a nuanced action within those Gen AI web and cloud applications. But in 2026, the game has changed. We aren't just securing people anymore. We're securing agents. That's right. Think about the scale. In what we now call the early days of AI in 2023, you might have had a thousand employees using ChatGPT to write emails. Today, you may have 10,000 autonomous agents that are running in the background. These are non human entities that are calling APIs, fetching data from your private repositories, and making real time decisions without a single human click. This is what we think about as the hidden architecture of the AI era. Traditional security solutions that are built for browsers and humans are effectively blind to it. If you can't see the communication, you can't secure the data. So today, we are releasing four new products. First, the Netscape One Ingenetic Broker that extends our existing AI visibility and security controls to public MCP based communication. Next, the Netscape One AI gateway, providing visibility, access, and runtime control into non user traffic that you want to inspect in your own deployment. So think of on prem or VPC that you don't want sending that traffic to the Netskope Cloud. Netskope One AI red teaming complements our AI gateway by stress testing those privately hosted AI models to ensure production readiness of your applications. And finally, Netscape One AI Guardrails, which extends our existing data and threat protection capabilities to stop AI specific threats and provide thorough content moderation with a single unified view of incidents. Let's first discuss securing MCP as part of our Netscape one agentic broker. This past December, we previewed our ability to deeply understand model context protocol, MCP, and apply visibility and control to all such agentic communications, performing data and threat protection. For those who aren't familiar, MCP is becoming the new communication standard, the standardized language that AI applications use to connect to enterprise data sources like Salesforce, Box, and Atlassian. Today, we're pleased to announce that we're moving from preview to general availability of our MCP capabilities, extending that same deep granular context we've always had for humans and applying it to this nonhuman traffic. We're effectively bringing zero trust to nonhuman interactions, allowing you to make informed real time decisions about what an AI agent can and cannot do with your most sensitive data. Bob, why don't we show the audience exactly what it looks like to have granular control over these public AI tools? Thank you, Melody and Jess, for that. Let's take a look at the Netskope One AgenTic broker in action. And we're going to start with a very popular use case. And here I have a Vibe coding application called Cursor, and this is using the power of MCP to connect an external source to bring in data so they can build an agent and bring it into an application. Now this makes it easier than ever. MCP is great. It connects everything to everything, facilitates sharing bidirectional, but it also poses a lot of challenges because there's no oversight. There's no controls in place. So there's a lot of risk. So let me kinda peel back the onion here and tell you about this use case. Here I am in the cursor application. I have an MCP server configuration, just a few lines of code by the way. And I'm an employee that works in HR and I'm building a application, simple application that provides a list of employees on medical leave and they can manage those employees. Some of their HR applications don't provide. So what this employee has done is they've used the version of this application to bring in this external source. In this case, it's from an application called Notion, and it's using the MCP server to facilitate that. So this is kind of the single source of truth where I have a list of these employees on medical leave that I pulled into this application. Now the problem is, once again, as I'm building this out, I'm bringing this sensitive data via MCP into these agentic transactions. So what we're going to do is we are going to bring in the Netskope One AgenTic broker that we've basically brought in to look at visibility, control, and protection of this AgenTic traffic via MCP. So the first thing that we do is we provide a look at what alerts we can detect the traffic. We can also govern activities based on MCP. So here's the universal policy builder. You could see I have selected MCP server as the category, and I can govern up to 22 different activities. I can also layer on different protections, like DLP, Threat Protection, and we'll talk about AI guardrails later on as well. So now that I've done that, let's look at the results. So I'm going to switch back into my application here that I'm building, and let's go ahead and run a query using MCP for that sensitive data. And as you could see, the queries are starting, the data is being retrieved via MCP, the Netskope One agentic broker sees that traffic and can detect that it's sensitive data, and as you could see, will block it in real time and protect the organization's data while not having to stop this user from using their application. And that is the power of the Netskope One AgenTic broker and the ability to provide visibility, control, and protection of these AgenTic workflows when MCP is the middle facilitating the transactions. Back to you, Melody and Jess. Thanks, Bob. Seeing that level of visibility into hidden agentic traffic usually gives CISOs a mix of relief, but also a bit of a cold sweat because these nonhuman interactions are not only happening across public AI implementations, but also in privately hosted environments. Netskope sees organizations building with tools such as Azure OpenAI, Amazon Bedrock, Olama Hugging Face. These are all in their private environments with the intent to meet data sovereignty and data residency requirements, just overall have more control over their privacy. We get it. You don't want your traffic to be sent to the public cloud, but what happens to visibility then? Yeah. Well, that is where the Netskope One AI gateway changes the game. We're extending our visibility and runtime controls directly into your private environments as well. Let's say you're a global logistics firm. You're building a private agentic system that optimizes shipping routes based on things like real time weather, fuel costs, sensitive customer contracts. Your internally built agents are gonna need to access high value internal databases to do that job. Well, this introduces the risk of an agent bypassing permissions or having an external prompt that tricks internal agents into leaking the contract terms to a competitor. But by deploying the Netskope AI Gateway, organizations can regain a central point of control. You can see and set guardrails around every single interaction between the agent and the LLM. You can enforce rate limiting so that a runaway agent doesn't suddenly rack up a 50,000 bill. And most importantly, you can apply context aware DLP and AI threat prevention to ensure that the agent only ever sees exactly what it's authorized to see. The Netscape one AI gateway does all the heavy lifting for you. It handles the authentication, the traffic management, the policy enforcement, all behind one single interface. And it even provides a full searchable audit log of all API calls for compliance. So now that we have a pulse of the AI ecosystem, we, of course, have to talk about the threat landscape. With new technology, we inevitably see new types of threats. For example, we're seeing a new class of jailbreaking where users or even malicious external agents manipulate a model into bypassing its safety protocols. The stats on this are staggering. Recent research indicates that jailbreak attempts exceed nearly 20% of the time, and we aren't talking about sophisticated nation state actors here. These attacks often take less than one minute and only five or six interactions to crack the built in safeguards of a standard LLM. These attacks are extremely simple to execute because the attacker is letting the agent do all the heavy lifting for them. Whereas before the attacker had to find a misconfiguration or a vulnerability to exploit and then slowly work their way into the system to exfiltrate data, now a well worded prompt, does most of the heavy lifting for them. And this is why it's so important to ensure AI applications are built right from the start. You wouldn't release code without a vulnerability scan. Why would you release an AI model without testing its moral and security compass? Yeah. That's why we're introducing Netskope One AI Red Teaming, now generally available. Before a model ever sees a customer or an employee, our platform allows you to simulate a motivated attacker's behavior. We actively try to trick the model into revealing sensitive data or bypassing its own guardrails. This allows developers to harden the model before it's in production, not find the security gaps after they're exposed to the world. But harden isn't invincible. You still need runtime protections, and this is where NetScope one AI Guardrails comes in. Yeah. Usually, implemented as three disconnected tools, we've actually brought together data protection, threat defense, and content filtering into a single unified view of events. So if a user tries to exfiltrate data via clever prompt, a siloed system might only see the threat but miss the data context. At Netskope, we present a single incident ID. So if a prompt attempts a jailbreak while simultaneously trying to export your source code, it's prevented and recorded as one cohesive event. We've also gone deep into LLM content moderation. We aren't just looking for bad words. We're using our own specialized AI to understand intent. We can help ensure that your enterprise AI isn't generating hate speech, citing copyrighted material, or inappropriate content that could lead to massive reputational damage. You can also set configurable guardrails, setting different thresholds for different groups. Your research team might need a looser filter for experimentation while your customer facing chatbot needs a zero tolerance policy. Yeah. We're effectively giving you the ability to say yes to innovation because we've built the safety net that actually catches the fall. Alright, Bob. Let's show folks how AI RedTeaming, AI Guardrails, and AI Gateway work together to effectively secure the entire I'm excited to continue with spotlights of key capabilities that help organizations navigate AI safely. And we're going to start with Netskope One AI red teaming. So here I'm logged into the Netskope console, and this is our red teaming capability where what we're doing is performing simulated attacks on your LLMs to look for cheek in the armor, if you will. And there's a number of different use cases that we provide. It ends up being thousands of prompts available as part of these different use cases that go from piracy and copyright to hate speech to weapons, etcetera. If I go over here, I've got two LLMs that I've configured, Bedrock, OpenAI, but you can bring in a variety of different LLMs. And then we perform test rounds on those by selecting the different use cases that you want to apply. Let's go ahead and look at the results of one of these, Bedrock. So here is basically the use case distribution, a number of past use cases. In this case, there's over 3,700 failed use cases. So these are prompts that got through whatever guardrails were in place by the LLM vendor. So this automatically shows some holes in the environment. You could get all of the details at the bottom. If I look at one of these failed cases, you get the explanation, the prompt that was used to put that chink in the get through that chink in the armor, and then ultimately the response that, triggered the violation. So that is AI red teaming. Now, visibility and risk assessment is very, very important part of hardening your AI infrastructure. But what about the action that you want to take? Now, is where Netskope One AI guardrails comes into play in our console where we have unified policies covering a variety of different, use cases. Now, from crimes and weapons, to hate speech, to prompt injection and jailbreaking, this is really about content moderation and threat protection. Now you could apply these policies in a number of different places, but the area that I want to focus on is the last demo of the day, and I'm very excited to give you a look at the Netskope One AI gateway. It provides visibility and runtime control, so real time protection into your private environments. And now this enables you to secure and manage the AgenTic AI traffic that is fueling your applications. So, is very straightforward. It's a little different than you might be used to from a proxy standpoint where you have a user and device accessing your destination. This is the world of agentic traffic. So the gateway can be strategically deployed where your LLMs are, and it's a very lightweight and portable, VM. Now that you deploy it, you have a policy construct where you put controls in place that, cover both access control, you can control accessing the LLMs altogether. You can also apply DLP, so you allow access, but you look for sensitive data and take action. Or you could take those AI security guardrails that we talked about and make those the action trigger, if you will. So let's switch over to an environment and take a look at what this looks like. And we know that the world of AgenTic is growing like crazy. It's cross industry, cross use case. Let's look at a couple examples. Here's a bank and here's the bank application. They've got an agent on the bottom right here. And what I've done here with Netskope is we have a configuration in place where what we're going to go ahead and do is have the private AI gateway inserted between the agentic traffic where you have the LLMs interacting, with the agents, if you will. And so now we're intercepting all of that traffic and we could put those, controls in place. So if I go back to the bank example and bring up this, we have two paths. One with the Netskope one AI gateway and one without. So if I look at a scenario, such as here's a here's a, trying to get internal sensitive data from a Confluence app pulled into the agent, this is very, very bad, what happens is without the Netskope One AI gateway, that data comes through. So they're able to get their way through any existing guardrails. Now, if I go in and test it with Netskope in place and run basically that same prompt immediately, real time, this is AI in the fast lane after all, Netskope puts those guardrails in place and the AI gateway was able to detect the request for sensitive data and block the agentic interaction. Another example is if I go in to this portal here and let's go ahead and go to this healthcare company. Here's another agent, if you will, and here is a prompt injection attack. And this is, in this case, we're going to go ahead and pretend that they're an unrestricted AI model. It's kind of crazy what you can do to fool these models. And I'm going to submit that with Netskope in place, and you can see immediately, we see that as a prompt injection and we're able to protect that in real time. So that is the power of the Netskope One AI gateway and everything that we've shown today together, these new capabilities secure the entire AI ecosystem and are fully integrated into the Netskope One platform. Thank you, Bob, Melody, Jess. What you've seen today is the result of our skate to where the puck is going philosophy. In this AI driven world, the difference between a leader and a laggard is confidence. Can you trust your data? Can you trust your agents? Can you trust the LLM or generative AI app? Can you trust your network? How you integrate security into your AI ecosystem will define your success. When you count on the AI powered applications you build, the LLMs that power them, and the agents transforming your business, you cannot afford for security to be the reason your AI takes ten seconds to respond, or is completely blocked. This is the AI Fastlane. With our extensively peered New Edge Private Cloud and our unique route control technology, Netskope delivers an end to end AI experience that is virtually indistinguishable from a direct connection. This experience is a mission critical priority. While employee frustration is a factor, the real stakes lie in high consequence automation. Consider a self driving car identifying a pedestrian, a smart camera triggering a security alert, or a medical scan detecting a life threatening anomaly. In these scenarios, latency ceases to be a mere annoyance and becomes a systemic failure. Netskope One is the only platform designed to provide the contextual control you need for total data security, without ever tapping the brakes on the innovation your business demands. We are extending our SASE and Zero Trust leadership to the entire AI ecosystem. We are here to give you the confidence to say yes to your future. Thank you for your time today. I'll see you in the AI Fastlane.