Inside the Enterprise Browser Rebuilding Security for the AI Era | Bradon Rogers, Island

0:00Can you talk about the risks that are developing with AI browsers and autonomous agents as they are designed by consumers or by users, but then are being integrated into the enterprise? There are a lot of agentic tools that are not being developed inside the enterprise. The agent is working over an application. There's hidden instruction buried in the application. The agent goes and follows those instructions, thinking there's the human giving them instructions in the process. The agent's nothing more than me hopping out of this seat and you see an empty chair, but the agent's doing work on my behalf.

0:34And we want to have policies that are built and let the agents operate in the confines of a given policy, no more, no less, and make them live in a place where they can be enterprise ready, no matter what the AI provider is that you're using. Automation enables teams to build and run on-demand agents that operate at speed across enterprise applications. How does Island keep track of what's being built by teams in an enterprise?

1:03I usually start by having you introduce yourself to listeners and give as much of your background as is relevant and how you got to Island. and then what Island does, and then we'll start talking about keeping the agentic world safe. I mean, that seems to be one of the focuses. Yeah, thanks, Greg. I'm Braden Rogers, Chief Customer Officer with Island. The role that I play is anything that's technology and engineering-centric that's in the field that touches customers. is the world I work within every single day. Island just announced some new functionality. Can you talk about that? Yeah, 1 ,000%.

1:54First of all, it's important to acknowledge the incredible innovation that we're seeing every single day and the breakneck speed with which everyone's experiencing AI. And, you know, it's everything from people leveraging generative prompts to exchange information and to make their work more effective. And it's obviously led over into some really interesting areas around automation and agentic workflows as people have engaged it. And I think it's important to acknowledge that many of the players that are doing this work are doing some incredible work. A lot of your different providers, the big brand names that everybody knows well that are the AI players.

2:39But a lot of the work that we see from that ecosystem is very consumer driven. There's nothing wrong with that. At the end of the day, they're chasing billions of users around the world. And, you know, but at the end of the day, when you're a large financial services firm or a large healthcare provider, you certainly have concerns with these things, you know, making their way into your environment. And then also the other thing is you want users to be empowered. You want to, so you want to, you want to wrestle with this consumer need, this consumer thing. and there's some goodness really there's amazing goodness with it um and you know we certainly believe ai is here to kind of kind of help us in the future not to decimate our futures and but we want to be able to harness that and leverage that power and enable our users very effectively and for us you know there's there's an there's an avenue for that's really focused around enabling the user but most importantly is allowing cyber security have a role in one of those rare times in the universe of cybersecurity of being a true enabler.

3:43Let's weave AI into the fabric of the user's workflows and let's do it safely and let a cybersecurity team play a very deep role in that enablement. And that's what we announced yesterday with a series of capabilities focused around the universe of AI that's square in the laps of everybody in the enterprise today. Yeah. Can you talk about the risks that are developing with AI browsers and autonomous agents as they are designed by consumers or by users, but then are being integrated into the enterprise? Is that the issue really, that there are a lot of agentic tools that are not being developed inside the enterprise but are being brought into the enterprise?

4:39Or is it even for tools that the enterprise is building inside the company? Great question, Craig. I think it's important to – I'm going to go up a level real quick before we go into the agentic side. because I think it's this series of AI resources. Sometimes they're agentic resources. Sometimes they're environments where users are engaging our corporate data into a prompt. Sometimes they're developmental tools with various forms of vibe coding resources that help us speed development of applications. But there's a whole series of resources people are bringing to the table for their jobs. and you know i think it's i think it's fair to say that you know your average user can find immense value in ai as a as a bit of a cheat code for the things that were difficult in their job may you know and you know sometimes it's about the knowledge that it can bring forth that otherwise they they may not have expertise in sometimes it's about automating their workflow sometimes it's building an app that you know now now a non-technical human being can create an application that's very sophisticated.

5:47But these things are coming at the organization so fast and so furiously that the organization is faced with, you know, just because these things have a lot of promise did not mean you abandoned your fundamental requirements. If you're a financial services firm or healthcare provider or any other technology or enterprise, I should say, it didn't mean you abandoned your requirements. And by the way, they're more important than ever. Your regulatory bindings that you're having to live by. Yeah, it's even harder now to live by those bindings with this universe of chaos coming to the table. Your data protection needs that you had, you want your users engaging the sanctioned stuff, not the shadowy universe of stuff.

6:26And so at the end of the day, it's the chaos. The orgs are trying to wrestle with the chaos. But then while they're wrestling with the chaos, not turn into the no police. because, you know, I think it's a very fair statement that a lot of the world wants to be AI empowered. A lot of organizations are thinking about how do we be forward leaning? How do we think AI first? And thinking AI first is not saying no to everything for your end users. It is shepherding and weaving in the sanctioned world of AI into their workflows and empowering them to use AI as their first thought process when they go to get into something that's creative or building something or being more effective, faster at their job.

7:11And so for my universe, or at least the way I think of it is, I want to think about AI empowerment first and AI mindset first for people that are thinking that way. And how do I encourage a user who otherwise might not think that way to naturally gravitate that way because it's now in their natural workflows to make their job easier, but it's also safe. And that is, there's a number of different angles in this whole thing. There is definitely the chaos that it's the onslaught of stuff that's coming at the organization that it's hitting them before they've got their sanctioned answers and their core strategy in many cases.

7:42Yeah, explain then how Island works with the enterprise. Well, I think it's important to highlight what we built at Island prior to AI. Just, you know, if you think about the workspace for the user, the majority of the work of a user is spent day in, day out using a browser. and it's a great piece of technology. It's one of the most important pieces of tech that has been around since the advent of the internet, obviously. And at the end of the day, the evolution of the browser, for our original use, was consuming information and then all of a sudden it became an application delivery. We receive our applications in that window.

8:22Yet the mechanics of that browser are not built for that purpose. They're actually built for shopping. I mean, you think about the targeted advertising and the focus around the analytics around the end user. It's a consumer-focused need, and that's okay. It's just that we said, look, there's some potential, kinetic potential, you know, pinup in that browser. What if we can uncork that potential and transform the browser actually into an application delivery platform where we build mechanics and policy to live locally for the experience of the end user? Things like, you know, building in data protection, building in contextual awareness, building in protection for the user as they, you know, engage the broader, you know, internet.

8:59But we want to use policy locally to govern that experience and then also make the user more productive, productive in a basic way, just as a starting point. They already know how to use a web browser. We don't have to train them. So what if you could just start putting mechanics of their work right at their fingertips in the, you know, again, woven into the fabric of the browser? And that's where we started the journey. And then, you know, along comes AI and its natural habitat as a browser. You know, for much of what the user engages, AI's universe is browser centric. So as a starting point, you know, then starting for us is weaving the AI mechanics into that user's workspace so the user has seamless access, but always bound by the policies that we had in play before, the data protection elements, shepherding them to the right tenants, you know, the right corporate copilot versus personal or the right, you know, Gemini environment versus personal.

9:50Or better yet, let them go to personal, but don't let the corporate data spill over the boundary, you know, into the tabs that are the personal tenants. So all of those are, you know, kind of fundamental mechanics that is what we launched yesterday is about the empowerment of the end user to take advantage and leverage AI and then taking it a step further is taking it into areas for agentic workflows, allowing agents to have the ability to operate in the workspace of the user, but have appropriate guardrails. These are all very important foundational things. And you've certainly seen stories. It doesn't take a lot of search through Reddit threads and things where people have experimented with these agentic browsers.

10:27You see analyst firms like Gartner publishing research saying, hey, be careful. Stay away from these agentic browsers in the enterprise for now. And there's a reason for that at the end of the day. We've got to put control around it and minimize the chaos, but also, again, back to empowering users. Well, let's talk about some of the new products. AI Protect, which allows safe usage of any AI application, consumer or enterprise. How does that work? I mean, where is that a plug-in to an AI browser or is it something more horizontal? Yeah. So if you think about how the user may engage AI, they're going to engage AI in their consumer browser, their Chrome or Edge.

11:14They're going to engage AI in an enterprise browser like Island. Are they going to use localized applications such as, you know, their local clients for ChatGPT as an example or whatever the tooling may be? For us, what we announced is the ability with AI protection is to live anywhere the user lives, whether they're working in the enterprise browser, if they're working in a consumer browser, using the Island extension to assert our policies inside the consumer browser, and then using Island desktop, which is a service that lives locally on the host to assert policy outside of the browser and govern the usage of AI resources.

11:48No, again, no matter where the user's working, govern those AI resources accordingly. That's exactly what AI protect is. AI protect is really two fundamental, very foundational things for Island, but very difficult for most status quo traditional providers. It is about having visibility, understanding who's using what, and building policies. And many of those policies, while they may seem very simple, like many upstream technologies have a hard time saying, yeah, let's let users use personal Gemini. And the company data is not going to spill. There's no risk of company data. No, what they do is the strategy, which you see oftentimes, is a strategy that's like you think about the pinch points, the upstream pinch points.

12:32We're passing traffic back through a cloud and inspecting at a pinch point. Their strategy has to be a block page. They've got to block the personal stuff. That's assuming I can even identify the tenancy accordingly. But we take quite a different approach to empower the end user. if they're using their consumer browser, they're using their enterprise browser, using the local clients, let them use personal stuff, but recognize when they're using the corporate resources and bring them into the fold, into the boundary so that they can have freedom of movement. And the agents fall into the same types of flows as well.

13:03Again, the agent's nothing more than me hopping out of this seat and you see an empty chair, but the agent's doing work on my behalf. And we want to have policies that are built and let the agents operate in the confines of a given policy, no more, no less. But again, make them live in a place where they can be enterprise ready no matter what the AI provider is that you're using. So for example, Slack is becoming agentic. How would Island handle that? I guess Slack is delivered either through the browser or through an app. This desktop functionality, how does Island wrap around that? that sort of thing you know an app like slack or any other app it's gonna live in the browser it's gonna live as a local thick app and the user engages those interfaces and those interfaces could they may be making web calls over here um the thick clients can be making web calls too because a lot of times thick clients are kind of browsers in disguise as electron apps but then many times in the universe of uh of of ai you also get into conversations that are you know other protocols, more modern AI centric protocols like MCP, you know, calls.

14:15And so, you know, for us, the ability for the user to use whatever client they're using, whether it's in the browser or a fit client and asserting our policy over those things, including MCP calls at the end of the day, so that when the agent is engaging something, the agents even doing it within the permissibility scope of what it should should be doing, not falling victim to, you know, I should go over here now and go to personal Gmail and start grabbing tokens from here or taking corporate tokens from this environment over here. And, you know, those types of things that we've seen, again, that have been proof of concepts over and over again, shown around a lot of that agentic world, you know, going amok.

14:53And again, it's early. So, you know, a lot of that stuff is naturally going to be experimental with. People are going to push the boundaries. We just want to make sure that the enterprise can take advantage of that stuff, you know, today and feel safety in doing that as well. So the Slack clients, agentic workflows, just like any other agentic workflows, whether there be a browser-based agent workflow or something outside the browser is really the scope of what, you know, we're focused on tackling. So the listeners understand the risks that you're addressing. What, give some examples of what can go wrong.

15:24First of all, I think it's important to point out that, you know, going back to the conversation of that, these, a lot of these things that's focused around the consumer need. So perfect example, you've seen examples of, you know, there's, you can search a couple red thread, you'll see examples of people exploiting the agentic workflows for prompt injections, for example. The agent is working over an application. There's hidden instruction buried in the application. The agent goes and follows those instructions, thinking there's the human giving them instructions in the process. And then the next thing you know, the agent's over here grabbing, you know, tokens, you know, corporate tokens out of a Gmail account that were sent via email, and it shouldn't be doing that.

16:03But at the end of the day, it's doing it and it's following what it thinks are instructions um we've got users that are very basic level type stuff users taking corporate data launching a personal app ai app and taking corporate data over to that personal ai app in a place we don't really want it because we just lost custody of our data because now it's now it's gone into that ai provider's ecosystem um it comes from a lot of different extension ecosystem you know you've there's i think i read the other day there's like 18 ,000 plus classified extensions that are classified as AI. That's a shadowy worm of things that wind up in your environment.

16:38You don't even realize they're AI living in your environment. There's another angle we've got to think about how we govern. So there's all kinds of risks of, again, it's a lot of it's centric around the data that you have in your organization, but in the agentic world, it's also centered around what the agent may do autonomously on your behalf, thinking it's doing the right work and inadvertently doing something it shouldn't be doing. So we want to build guardrails and policies around those agents, just as we have around the end users. Again, a very smooth experience, but we want to let the user also, by the way, because the agent doesn't exist independent of a user.

17:11User sends instructions. We want to let agentic stuff be presented to the user to be able to take advantage of these things and not have to worry about the risks themselves either. How does it work under the hood? Because yeah yeah 1000 so anywhere you assert an island policy whether in the browser or whether you assert it inside of the extension or inside of island desktop again you think about the movement of of an agent the movement of the agent is moving in a very similar way akin to the end user i'll use an example yesterday i was doing some agentic work and i asked it to go to a social media platform and find and connect me to certain people that were in the sphere of a particular topical area that I wanted to learn about.

17:56And I let the agent go and do that work. At the end of the day, the agent would be bound by a policy that lives in the browser in that given example. So if the agent decided, oh, it saw some instructions to go over and take this data from here and go over the boundary to the personal stuff or some other area it shouldn't post the data to, the policy lives in play and keeps it from doing that. So the agent would still bump into the guardrail that was there for the user. And maybe we may even have more rigid guardrails for the agent for certain things as well. So the guardrails could confine the agent to specific sets of work, like only automate this one thing, but don't go out beyond the confines of that.

18:33So policy lives locally. That's really important. Lives locally in the browser for those things. Lives locally in the extension, asserting itself in the consumer browser, and lives locally in the host with Island Desktop. Again, those were all announced yesterday as well, part of the Island Enterprise platform. And then so as a result, when the agent starts asserting its will, oh, let me go touch this thick application outside the browser or touch this folder in the file system. The Island desktop is asserting its policy for the agent, just like it would a user in that particular case. And then when the client is engaging some MCP resources as well, we assert our will in that process as well.

19:08So that again, anything going into the MCP flow or coming out has been governed or at least audited at the appropriate level for those engagements. Audited and logged so that there's a virtual paper trail of everything that goes on. How is it configured? I mean, does the enterprise, how do you work with an enterprise? Yeah. Well, I think one thing that's important to point out about what we've built for the past five years, it's a very novel approach to asserting our will, our organizational will. By the way, the will is not just cybersecurity. security it's our will for productivity resources for the end user and automation resources for the end user and digital experience and stuff like that but it is a very novel approach you know you think about it you already know how to use a browser today craig if i had to onboard you in the universe of island i would tell you to go to the app store of your device of choosing or go to the island download page and download island and something tells me somewhere in your past, you've actually downloaded and installed a browser probably many times in your life.

20:16You launch Island, you log in. The org has built a policy that the moment you log in, the browser and our services instantly reach up to the cloud, grab the policy that the organization built, and boom, all of a sudden your application entitlements are provisioned for you. The policy's provisioned locally, the automation resources, et cetera, are all provisioned locally. So for you to onboard yourself as an end user, it's a self-provisioning exercise because you You need to know two things in the universe of Ireland as an end user, especially if you're self-onboarding. You need to know how to go to an app store and download an app, in this case, our browser.

20:49Second thing you need to know how to do is log into the single sign-on provider for the org that you work for. If you've ever logged into your ENTRA ID or Okta environment, you're done. All of a sudden, context, based on the positioning of the policy that the org built, is now staged for the end user. So the end user's experience is literally nothing more than engaging the familiarity of a browser. And this is why we don't require any training for the end users either because they just now see it and they go straight to work instantly wow and so island can can find and read all the the policy documents required to to govern uh agentic uses and that sort of thing yeah so you obviously the org is going to build a policy at a time the policy is going to have all this and by the way many of our island customers already have these data protection policy they've been building them working with island for years and deployed at massive scale so they built these things.

21:41And so now they've already built them. Now AI comes into the fold. You simply empower, enable the capabilities to integrate to your proper AI providers. Now AI is, the AI is elements are living slipstream right into the existing policies that the user already has. And as we continue to evolve capabilities, we'll always just evolve and begin to slipstream straight into the capabilities of the end user. And that includes even new apps in the AI universe. One of the things we announced yesterday was this area around AI publish. You want to be able to leverage resources like Vibe coding tools like Lovable and Cursor and Cloud Code, stuff like that.

22:21You want to build those apps and be able to deliver them and actually have them run and then have those apps slipped right into the island policy as people deploy the apps and deploy them at the same speed that they're deploying them in the Vibe coding tools that they're creating. So the speed is matched for deployment that is happening within the delivery. but cybersecurity policies are never abandoned in that process because it's literally slid right into the exact policies you already have because again that world is very much even though they're ai-coded apps they're web-based apps they're they look and feel web-friendly you can build thick apps but you know most people are building web-centric type apps if you know they're very rich type stuff and uh again you don't have to walk away from all the stuff you built in the past that it was important for you.

23:03Yeah, but this is a B2B product. It's not B2C. Or can it be B2C? I mean, you were saying that you go to the App Store and download the app. How do enterprises use it? Do they have a, is it by seats or, yeah. Yeah, it is definitely an enterprise focused. And I say enterprise. It's not just for the large end of town. It's for the small end of town, too. So we've got orgs that are small orgs that are getting value out of Ireland because that small org, let's say you're a small bank with 200 employees. You have the same regulatory bondings that a large, massive bank has and has to live up all the federal government requirements and maybe international requirements as well.

23:57You still have that, but you can't afford the same stack they have. You don't have the same practitioners and resources, so you need to do more with less. And we empower that for the smaller org. So it's an organizationally focused effort, not focused on consumers. Its focus is around, you know, user-based pricing at the end of the day. So it's pretty simple. You know, at the end of the day, a user can use as many apps as they want. They can use as many browsers, as many devices. That's not a consideration. It's not a concern. So we just make it really simple for people to be able to consume at the end of the day.

24:31But let's say an organization with 200 people, do you charge by seat and then you deliver an app for each of the 200 people? Or each of the 200 people downloads the app on island on their enterprise's license? How does the organization get Island into its employees' hands? Great question. So it goes one of two ways. I used the extreme example earlier when I thought about Craig going to the App Store and downloading the app. That's often what we see in the unmanaged real estate, a third-party contractor, an unmanaged device, a third-party BPO, business process outsourcer. You've outsourced your call center to a third-party entity.

25:21because you don't have the ability as an organization to force install software on somebody's machine you don't own. That breaks the laws of physics. So we give an easy way to onboard for the unmanaged, but on the managed side, you deploy like you do any other software, leveraging your existing software deployment methodologies that have been in play for years. So we just slip island into those processes. And oftentimes in many of the orgs were widely deployed. We're just part of their core build when they ship a device to somebody. Island is just part of the foundational build for those people in that case.

25:52And then the onboard themselves, they just go log into the SSO and they go to work. Automation enables teams to build and run on-demand agents that operate at speed across enterprise applications. uh does uh how does uh island uh keep track of what's what's being built uh by teams in an enterprise or is it as long as you're working on an island desktop or or in an island browser everything is automatically covered or does somebody have to submit a new agentic capability to island to be reviewed? There's a couple of ways to go about that. the most effective way is to leverage AI automation in the construct of provisioning workflows for your end users that are enterprise-grade workflows.

27:05The thing you want is you want predictability out of your workflows for your end users. You want to take, let's say, a call center worker, and you know that it takes them 10 minutes to accomplish this task when someone makes a call in for a reservation or for getting an account status. You don't want them to have to go through and learn that process. And sometimes those processes could be across three or four different apps. They got to learn. You hire the new, there's the onboarding experience that they got to learn the process. And especially if you're outsourcing to a third party, it's even more painful.

27:39It wouldn't be great if they could just push a button and it goes and accomplishes those tasks. So a big part of what we do is pre-staging the workflows that we publish. Publish those workflows to the audiences of users. And then when the user visits that part of the application, serve up reminders based on the learnings we have about that user. Like, for example, one of the things we're really heavily focused on is learning the role of the end user, understanding the nature of the end user. I use a parallel. You know, when Elon Musk, you know, has gone down the path of building autonomous vehicles, he didn't put, you know, inspection points at every intersection and watch how cars drove at every intersection.

28:27That's how the existing world of technologies, you know, when you, especially when you think about the cyber universe, would approach the problem. They would, you know, they get the pinch points, the upstream cloud proxies, and, you know, just everything's being zeroed in on a specific pinch point so it can explode outward. It'd be akin to Elon Musk trying to learn how people drive cars by putting cameras at every intersection. Didn't do that. They sat in the seat of the car. They observed how traffic works as you drive down a street because not every street is an intersection where a camera would be in the pitch point example.