Island CEO: Quantum Changes Everything in Cybersecurity | Apex Series

My pleasure. My pleasure.

Well, I think each of those companies you mentioned are kind of taking a different view of what they want to do. Some want to be an extension in a browser. Some have a need to have their own. It doesn't matter. You know, when we built the browser, we had a browser in the first two weeks of our existence. Having a browser doesn't mean a lot. That's easy to do on a Chromium open source project. It's the 200 engineers plus that have been building DLP, networking, credential management, digital experience, anti-tampering. I mean, the list goes on and on and on. So, as we see these new incumbents, in many cases, like in the OpenAI case, we'll make sure our platform supports them. We'll make them better. Great. From the competitive cyber security guys, we've never believed the secure browser was all that great an offering. If all you do is improve security a little bit, it's not worth swapping browsers out. Without a doubt, we have an extension for that, right? And you can just use that. When you really say, 'Okay, now I want to improve the way the world works. I want to reduce the technology stack that it takes to support a knowledge worker. I want to make a knowledge worker more productive and I want to do so securely.' That's a real effort and you don't get that with a hundred million purchase of a company doing a million or two of ARR. That is a long, long effort. And we still stand alone in that. You know, we haven't lost a technical head-to-head in years. And so it is a different game. Doesn't mean we get to every customer. There's a lot of customers out there, but the world has expanded that enterprise browser. Now, what it really means is all compute on the endpoint. We're outside the browser. We run on everything that runs on that laptop of yours. That's the space we now operate and govern. The world does expand.

It wasn't that big a surprise to be clear. AI was coming. One of our investors, Georgian, has an entire AI discipline with an AI lab that they had from the first dollar they ever gave us working with us. Its explosive size was a surprise. We like, we didn't see that coming. But AI being important in a fundamental premise of the future, I think a lot of us saw that. We definitely saw that six years ago. Not at the scale that we are like envisioning an Anthropic was not something I wish I could have seen, right? But definitely did not. So what it's done for us though is it's allowed us to take this wonderful thing we've done. So going back to the core premise of the word 'enterprise' for us, the reason we use the word 'enterprise browser' not 'secure browser' or just 'browser' is it's the knowledge worker, the IT stack, and the security. So now you think about AI. What is AI? It's an agentic worker. It's an agentic capability. It needs capabilities. It needs an IT stack. It needs security. And it turns out what we built is exactly the premise that needs as well. So that allowed us to go into the AI Protect offering. So you want to have AI, how do you protect it in all directions? How do you make sure AI doesn't do what you don't want it to do? How do you make sure your users don't use it in a way you don't want them to? How do you control that? That was interesting and that's a compelling part. But the mistake cyber security often makes is we focus on that last part. So now if you want to find a buyer, it has to be a company that has rolled out AI, is grappling with security challenges, and now looking for an answer. We saw let's move our offerings forward. So we start to build things that help AI be successful to the knowledge worker. So we built AI Publish. This is the idea. You can take that Replit, that Lovable code, anything, and literally push it down to the user in a safe way. Put it in a container. Now it can run. We can move things quicker. So now a vibe coder over in accounting can push out an agent inside an important organization to everybody in it without risk where normally that security review, pipeline processes, all the things that don't screen next generation. Then the idea of how do you inject AI into existing processes is part of that presentation layer. You've got an existing application. You want to have AI play a part in some of it. You're not ready to rewrite the whole thing. How do you swap it out? So we at the presentation layer, we can take that button to maybe create a new customer and call a whole AI sequence that does the work for that with a headless browser off to another environment all while you manage that process. So again, how do you get it out? Then there's the whole access to train. We see the end users in a way nobody else does. We know everything they're doing. So much like you've seen on news where you see people wearing the robotic hands working in a factory to train a robot. We can do that digitally for a call center, for a front desk workforce. All of those things we can capture in a way AI loves screenshots with enriched data so that it can learn, grow, and thrive. So now we're being pulled forward to companies say we want to embrace AI. We want to do more with it. Let's get Island in here and oh yeah, we're going to protect it, but we got to get it out to everybody. We need everybody to use it. We need to learn. We need to train. So that's been this concept of that word 'enterprise' and not just security, not just end user experience. The idea of the triumvirate, it is amazingly viable in the AI generation.

I think it is very viable to have a safe AI interaction. You know, one of the things that frustrates me and we've dedicated Island to approaching this differently. At Island, we're using AI to win. AI is not the boogeyman. It's not the bad thing. Mythos was the first touch of AI to cyber security. And that was unfortunate because, well, it's an amazing product, don't get me wrong, but it was unfortunate because it exposed vulnerabilities. It exposed attack surfaces that we didn't want exposed, right? We wish weren't out there. That gets people thinking about AI and then they think about data upload and all that. And it's typical old cyber security approach. Mobile phone shows up. Oh, how can that be bad? Instead, in this case, we're going it's the answer. You know, put physical analogy. I think you would agree if there's a police officer out in front of every door in a city, that city would be safer. You can't afford it. If you could monitor every traffic light and understand everything going on, you can help traffic patterns and solve easy problems, but you can't. But with AI, you can. And so, we want to be the leaders in bringing AI to solve the problem where AI is the answer, not the problem. Not the thing you have to protect against, not the thing you have to put rubber, you know, guardrails around, but how are you going to deliver policy-less protection at scale? To do that, you need to see everything that's going on, not reduce it. And that's a fundamental change in cyber security. Every product built till now was about reducing data. If I'm going to go to a SIEM, all this data comes in. The money was spent on how do I collect the data and then how do I shrink it so I can see the one or two things I have to pay attention to. That's not what we want to do with AI. With AI, we wanted to see the whole thing. I got to sit through an amazing presentation by two companies and they were hired to essentially build AI drone pilots. And the first way they did it was they trained them like an Air Force pilot. They put them through the Air Force pilot training process and then they flew missions and it worked. Another one they said don't train it, let it fight itself, let it go. That model was far superior. This is what we have to do with AI. It is not trying to get it to take the way we used to do it and force it now to be done in some automated fashion. We want to give it the data, we want to let it run on those IOCs, we want to let it see the world and come back with its findings. And if we do that right, we could fundamentally have the first dramatic systemic shift in our ability to defend ourselves. I think in the next 5 years we will fundamentally change the risk of being breached by embracing AI and all the talk right now is how AI is the boogeyman. It is not helpful. We need to go to the next stage.

Listen, they found zero days in some very important technology. They found stuff that impacts national security in a real way. To write it off as a marketing stunt would be unfair. Now, did they play it up for the benefit of Anthropic brand? Sure. Did they roll it out in a way to monetize it really well? Yes. Good on them. That's what they're supposed to do. But they did deliver real tech that matters to the world. And to be frank, it's going to come from a lot of places. That's the crazy thing about these models. I don't think we fully understand what their mode is yet. And maybe they do, but I don't think we do. And so why won't there be another model that maybe finds different things? And I always take AI back to very much like a human, right? We've got brilliant experts. You know, Einstein saw one part of the world of physics that another person didn't see. Somebody else saw that. I think we'll see multiple models running with different contexts given to them that discover fundamentally amazing things for the betterment and the detriment of organizations. Mythos is the first taste of that. It is the first taste of it.

So I agree with the theme of your question, not last line, first line. And here's how it's working out. So some of the largest banks in the world are doing this already. They're sitting there going, 'Okay, I can't patch at a speed appropriate to shut down what Mythos is finding. I'm in 80 countries and I run the world's GDP on me. Literally banking countries can work together because we exist. If I mess that up, we grind countries to a halt.' So the idea of rapidly patching just because a vulnerability was found impacting the availability of those resources is just not something they can grapple with. So what do they do? They're going, 'Okay, we have to fix that. We will fix that. We will get to those. We will solve those. We will do that. But we don't want to have to knee-jerk reaction to that. So, we're going to look to our perimeter. And at the perimeter, can we apply a virtual patch, a compensating control that will allow us to live with that vulnerability for a little bit?' The analogy I like to use is the lock on my office door is broken. Let's go make sure the lock on the front door is right and make sure nobody can get in that's not an employee. You know, it doesn't solve everything, but it does give a level of acceptable risk for it. So Island from the start we built it to be fully API-driven. Everything that happens in the browser is API-driven. It's very unique to us. Everything can be driven by direct code 100%. So that allows somebody to go, okay, I'm going to build an automated system with AI that tells my perimeter how to reply to this vulnerability. What can I do on the perimeter to do that? And that's a beautiful thing. You could take Mythos which found a vulnerability we don't know how to apply another AI engine that says well how do I limit that with these perimeter controls and it can respond. So when it responds, if it sends out the commands to that and you can accept those commands, all of a sudden you are a fundamental part of the infrastructure to do that. And in cyber security as well as probably everywhere else in software, all that will remain are primary data providers, the things that produce data, the infrastructure that it takes to run our environments and things that are part of the AI sequencing chains, whether it's the AI direct or the AI response plane. Island were all three and it's a very unique place to be and I wish I could tell you when we sat down six years ago we said let's build something like this and came up with the idea of Island but it's just evolved and the world has shifted to the value of the end point is so much more important today than it was 3 years ago and being the disruptor on that endpoint is an enviable place to be and we'll see challengers and people chase us but it's hard work they're going to have to invest a lot billions to catch us.

What a great question. I would say quantum, the quantum armageddon as it's... And I hate being the fear-mongering guy, but I don't believe this is fear-mongering because it's just fact. The timeline moved up. It was 2036. Now it's 2029. And what that Armageddon day is is when will quantum compute be available to adversaries that can break the encryption and basically see inside our most important conversations. So the way we've thought about cyber or the way we've thought about security, we encrypt our communications. We encrypt our communications. We don't block who can get in the middle of them. That pipe that runs across the ocean, we know people can see inside that, but can they unencrypt what's in there? And we've had confidence around our encryption levels protecting that. When Quantum's available, it rips through that encryption and turns it all into plain text. So now we'll see banking information, we'll see critical information, we'll see secrets and passwords and everything else is now available in clear text. That is horrifying. That is true technical Armageddon. That's a lot scarier than Mythos showing up. So you have to start to say, okay, we have to engage in quantum encryption. That's not that bad. We can live with that. But what does it do when my browser spins up a connection to a server somewhere else at a quantum encrypt level? The idea that you're going to get in the middle of that stream, break it at line speed, look inside of it, close it back again, and send it on its way just doesn't measure up to common sense. So what does every large network maker do next? How is their gear? Because their gear got valuable because it moved from switching and routing to insightful information. Well, it's only insightful if it can see the traffic. It's not insightful. It goes dead soon. So, you have to move all those controls to pre and post encryption. Who's pre? We are. Who's post? Amazon, Google, Microsoft, those things. That's where the controls now have to go. Sitting in the middle is an antiquated thought process. And you think about almost every major cyber company has a very important set of product lines that does that. When that disappears, their EPS hits, the value of those companies drop, and they're going to find a fun game ahead of them. It's a very interesting time in cyber to figure out how quantum works and unfortunately the word quantum got associated with, you know, sci-fi and never going to be here and it's, you know, so I think a lot of people wrote it off. Meanwhile, those in the know, it is real and it is coming and it is coming fast. And then just to kind of round out your question, if you take AI into quantum compute with a self-learning capability, good night. It'll be really interesting to see what happens next.

I think it does transform but I think the CISO community, you know, cyber security has always been this weird thing. I didn't start from cyber and when I joined it I was behind the other cyber guys by about a year. Your knowledge in cyber gets you about a year ahead of somebody not in cyber because it changes so fast. To be good at cyber, you're constantly learning. You're constantly staying on top of things. And so the CISOs we have now that are good, the really good ones, they have made a career of learning. They're actually some of the best positioned people to take on this next generation topic because they've been learning. Many of our other roles in IT have been able to do the same thing for a decade and so they might be a little out of practice of reinventing themselves and relearning where our CISO community by default is always on the cutting edge. You know, CISOs meet with startups not because they get talked into it because they want to understand the cutting edge of everything because they know that's their future to deal with. That mentality sets those up. The CISOs I work with are going to become CIOs. They're going to become the leaders of this next generation because they have been able to learn as a fundamental premise. That's how they got their job. Constant learning. And that's where we're at now. What you knew about it 12 months ago is almost irrelevant. It's what is the next 12 months and more importantly, what does the next 24 months look like? So all of us have to reshape ourselves fundamentally and that gives opportunities to those of us that can learn quickly, can assemble information and put it to use and it puts challenges for those that have gathered their information over a slow growth period. It is something in security we've had to become great at and the best CISOs are.

The token spend, yes, it's gotten out of hand in some organizations. Token optimization has not shown up yet. I built an agent recently that consumed tokens. I enjoyed what it did and then met with one of my developers. We tweaked it a little bit and all of a sudden it didn't even consume a token. I wasn't educated into the ways to do that. I send Claude 20 PDFs on topics I want to go through. I don't run it through a markup language creator, right? I don't swift it. If you do that, that's 70% less token usage. We are all learning how to do this and do this right. The aggressive spend is, and I do this in my own work, I'd rather you using AI right now and fighting your way through it and wasting money on it where we could have optimized it for you than have a slow roll out because unlike going to cloud, unlike going to mobile which were other big changes, this one is all about competitive. My customer didn't know if I was on Amazon or my own data center. It didn't change my economics, it didn't change my ability to compete. This does and in business winning comes first, optimization comes second. These companies are trying to win, they'll optimize next. It's a natural path. On the data exploitation, the discovery, it's actually pretty easy to manage at the end of the day. The models work with you. If you have control over your last mile, which is the challenge, and most people don't, but if you get control over your last mile, you can make sure corporate information stays in closed models, stays in your models, doesn't go to public models. You can do things of data abstraction and all these other items to it and AI can help you with that. Most of people's interference or experience with AI is a single experience under a single outcome. Take how we do development. Our development here, 90 plus percent of our code goes through. That doesn't mean we sit at Claude and Claude writes it and we say thank you. We write it. We run a judge against it. The judge tests it, validates that it's there. We run another one to make sure there's not a security issue, all three models have to be in sync and work. And they're all three independent versions of that. So, while an agent can go rogue if you build it correctly, there's another agent watching its behavior, watching what it's doing, and is it living up to it. So, when you put through a real pipeline, not you and I playing on our desktop, but a real pipeline to do an agent, that agent will produce a dossier, a statement about what it does. I built an agent to look up who you know in Salesforce who has what account, what are they spent, what are they doing. Okay, when you build that agent, great, now the agent should tell you I only need to connect to Salesforce, I need to connect to this Salesforce, this is the data I need, here's where it goes, here's the networking required and here's the expectation digital experience. Great, if you have that then you go put control points around that, only go here, I'll give you the credentials when you go there, you don't need to know them so you can't go anywhere else, I'll control the data boundary. So you set all that up and you naturally have a very efficient, thoughtful, safe environment. That's the reality of the future. Not I wrote a prompt, did something silly and that goes through. That's our moment in time of learning how to work with this, you know, amazing technology. We're not writing prompts. The front desk workers at Marriott aren't writing prompts in the future, right? Your call center workers aren't writing prompts. Honestly, most of us aren't. It'll be a select few that do and the rest interact. And that's where the economies of scale come in. So I think most companies are starting to come to grips with this. It's security's job when new tech comes up to scare the [__] out of everybody. That is not what we need to do here. This is manageable. We can handle it. Yes, you do need control the last mile. That's what we do. Great. It's not hard to deploy us. It's easy to get out there. But once you have control over that, AI can be just a wonderful playground to try, grow, and essentially compete.

I think this idea that we need to solve a problem is one, like we've got some cyber security going, 'Oh, I need an answer for Claude Code' as if that's all they're going to deal with, you know, and instead of asking the bigger question, I need an answer for data control for this layer. I need an answer for network control at this layer. Claude Code is your first taste, right? It would be like the first SaaS app being Salesforce and saying, 'All right, I got to figure Salesforce out.' Like there's not 95 others coming six months later. So when we think about this, as powerful as Anthropic is and as great as they are, even Anthropic has multiple models. Even Anthropic has multiple stuff, right? So we're going to be in a world where we're going to need to have dexterity of control. We're going to need to be nimble and we need to think about our control layers not at a response product layer but at a functionality layer. So, it's almost like we need to go research Swiss Army knives, then how do we cut this one thing? I think that might be the area some miss out on. Maybe they don't miss it. They're just reacting to what they have to deal with at the moment. They know it. They'll get there. But that's probably the biggest biggest hole. And I do think some of it's overstated as far as the hallucinations and things like that. You know, hallucinations are hilarious to me. The number of bad decisions made by humans is so great. The fact you can trick an AI into making a couple bad decisions and you want to doom the product for it cracks me up. It's all context, right? Bad decisions, just like with humans, are made from not enough data, too much data, or bad data. Same goes with AI. And you know a basic test that you should run that most people aren't is you go in, you build your model, you do your activity, then destroy it, rebuild it. If you get the same outcome, you know you're provided the same context. If you didn't, you have context coming from somewhere you didn't know. Owning that context layer and making sure you have an audited context layer that you understand why and what it's for and why it's there. I think that is something that cyber will have to start to grapple with but that will be part of the art form of AI long-term viability and greatness.