Zscaler's Jay Chaudhry on Zero‑Trust for the AI Era

Many people plan entrepreneurship. They have a journey. They have a plan. For me, I may call it, I'm an accidental entrepreneur. Having been born and raised in a tiny village in the foot of the Himalayas in northern India without electricity, without water. So, I had no idea what companies and entrepreneurship is. But I was always a hardworking good student, got into engineering and came to the US to do my masters. So I thought I'll be a software engineer for most of my life like most engineers. The first time I got into entrepreneurship was sitting in Atlanta reading about this company called Netscape that had just gone public around 95-96 time frame and reading about this young chap Mark Andreessen and say wow this young guy is starting a company that's going to change the world where you could access information using a standard browser no special software that should be very powerful because it's wonderful. But then I said if that happens everyone connects to the internet which means someone has to worry about security, cyber security and that may be an opportunity. And then talking to my wife I was saying if this young guy could do a startup why shouldn't I do a startup was the first time idea coming out of nowhere and it just built upon it and that's how I moved away from being a business leader in a corporate world, corporate America to be a startup guy.

That's right. So the first company was SecureIT and for that company I failed to raise VC funding because I had no experience in any startup world and since I got passionate about it. So my wife and I talked, we put our life savings on the line to get the company started to make it more exciting and my wife quit her job. So there's no turning back. Burn the bridges behind was the notion. The company took off. We're very focused. We're very good at what we're doing. So the business took off. That company got acquired in less than two years and everything went so well and I said, 'Wow, startups are supposed to be hard. This was easy. This must be a fluke. So let's do it again.' And also you know you want a bigger challenge in life. And then we said this time we do not need to talk to any VC. In fact I said we could only put so much money in one startup. So we should do multiple startups but do them like raising having kids. You stagger them one or two years apart. So we ended up starting three companies: one e-procurement ASP using Ariba software that was a precursor to SaaS, single tenant application software provider, an email security company called CipherTrust, and a wireless security company called AirDefense. Good luck, good timing, great teams, they all became successful and eventually got acquired by large companies. That took me to 2007-2008 time frame. At this stage, I had no interest in doing one more startup and selling it. I wanted to do something big, something lasting. And that's where I was looking for what could be the long-term opportunity where we could build the business. And that's where the ideas of Zscaler came in. Sometimes I got told, 'Gee, how did you come up with that idea? Must be some very creative thing.' Not really. It was fairly common sense. If you think about 2007-2008, internet was a big source of information. And I have been using Salesforce and NetSuite since year 2001 when each of them was under $10 million in annual sales. But I had used them. I thought SaaS is the way to go. SaaS should take off. And then AWS had just launched infrastructure as a service, compute as a service, storage as a service, which made so much sense because why not do compute as a service? And then Apple had just launched iPhone with a big screen. And I said, if we all become mobile and applications are everywhere, are you going to keep on building these moats with firewalls? It'll be a crazy idea. So, let's think outside the box. Let's kind of flip the security thing on its head and not do firewall and build something totally different like a switchboard, a policy engine where everyone's untrusted. The network is merely the transport. You don't do network security. Those were some of the ideas we thought of and that's how Zscaler was born.

Absolutely. So the firewall based security is generally called castle and moat security. If you need to protect your castle, what do you do? You build a moat around it. So bad guys can't come in. And then there's one place to get in and that's the moat has a special door and they check you before they let you in. And once you're in, you're supposed to be a trusted good party. Now, that model worked fine for the traditional world of computers. Here's your data center and here are some devices sitting there and you have a network connecting them and you want to make sure that if you connect to the internet the moat is a firewall. It's like a door. It's like a gate keeping the data center safe. And then you had to move to 50 offices. Now you're connecting your data center to 50 offices. It's almost like you had a big castle, then you have 50 small castles, then you connect the big castle to 50 small castles through tunnels. But once somebody gets inside the castle, they can reach everywhere. That's the problem of the network security, firewall security model. You're in, you can cause lots of havoc. Now, what firewalls tried to do was, 'Oh, if you're worried about that, why don't I create more walls? I have wall here, wall here, wall here, wall here. I'll sell you lots of firewalls.' It creates a big problem, maintenance, all that kind of stuff. That's the model I was familiar with. In my first startup, SecureIT, I was actually designing, architecting, selling, supporting firewalls and had seen all the limitations of it. That's why I wanted to bring a disruptive idea also. You know, I'm a big fan of reading about technological changes. I've always believed that technology incrementally changes all the time, but every decade or every two decades there's a step function change. That's what drives innovation. That's what drives business productivity. Give you a couple of examples. There used to be an analog world which did pretty good. Then we had to shift to digital world, digital technology, much better, much cleaner. In the world of software, we used to have application in our data center from vendors like Siebel Systems and the like. Then came Salesforce. It says why do you have to keep on worrying about buying servers, deploying all this stuff? Connect to me like a utility service and use it. So those are step function changes. When I looked at the world of IT, I said applications are changing but nothing has changed the world of cyber security. Same old 30-year-old firewall technology. We need to do things differently. Let's not trust anyone. At the end of the day, users need to access applications. How do they make sure the right person can have access to right applications? So, I thought of a simple metaphor, an old school phone switchboard. Years ago, if you want to talk to John, you call the switchboard and said, 'Please connect me to John.' You are connected. You had a great conversation. Then you said, 'Oh, I want to talk to the president of the US.' The switch said, 'Sorry, you're not allowed to go away.' It's really connecting the right party to right party. Such a simple elegant architecture. I like simplicity and innovations. The way I also thought was the service should be simple. We all used to love DVD players. You recall those DVD players? They were wonderful. And then Netflix came around and say move over. You simply click on a button, you get what you want. That's the kind of service I envision for Zscaler.

Yes, I have it in my museum room.

Yeah, so I have often said AI is wonderful. AI is powerful. AI is dangerous. Okay, it's almost like any new technology. A knife in the kitchen is a wonderful tool unless it's abused. So, here are some of the bad things AI can do that makes it very hard for companies to protect themselves. When an attacker attacks, they find a target. It's like a bank robber has to find which of the 500 branches do I want to break in so they can break in without getting caught. In the cyber world, they find our public IP, my firewall, my VPN, my application portal. It used to take weeks for hackers to find some of those cyber targets for your company. Now you can submit the query to ChatGPT and it can give you a beautiful table format of all the firewalls with vulnerabilities in them. Makes it so much easier. And then often people will do phishing emails and those emails sometimes could be typos, missing spellings, coming from Nigeria or wherever. Now ChatGPT can do beautiful phishing email with the writing style of your CFO and coming from your CFO. That's hard. Makes it hard. Three, once the bad guys get on your corporate network using stolen credential with VPN or something like that, now you can use automation of AI and AI automation can allow you to find mission critical applications, encrypt them or do whatever needs to be done and ask for ransom. The Anthropic attack you talked about, it's essentially literally hijacking an agent. Okay, agents are somewhat like people. People have been the weakest link. I believe in future agents will be much weaker link because the code can be changed. Things can be hijacked and no agent on your network has access to all kind of application. They can try to do bad thing and the zero trust architecture we pioneered at Zscaler actually plays an important role. The zero trust architecture as I described the switchboard says everything is an island, everything is untrusted. You come to a checkpoint, you get connected to the right party. If enterprises embrace zero trust architecture these hijacked agents or users whose credentials have been stolen they won't be able to get past maybe a given application beyond that. So it's fundamentally very powerful architecture to safely embrace AI security. So we and there are other aspects of it but starting with a zero trust where agents are only trusted for a given amount of time to do certain tasks and no more is the right way to do these things.

Yeah, that's a very good question because our customers want to embrace AI, use AI, but they're all nervous. They want to do it securely. The bunch of new startups all pitching their point products and some of the large companies who have bought some other companies acquisition here and there but lacking integration. So what have we done? First of all, Zscaler pioneered zero trust exchange for users. Then we took it to cloud workloads. So workloads can talk to other workloads in a zero trust fashion. And then we made zero trust branches where each branch is like an island like a cafe. So one infected machine branch can't infect others. And now we are extending our exchange for agents. We call it agentic exchange. Any agent being able to talk to any application service will go through our exchange where we're going to start with authentication, identity, authorization and a few other checks to make the right connection which is natural for us to do as we do exchange for users. We want to do it for agents. Important. Now the second question there ends up being if you want to do it who provides identity of these agents? Each agent must have identity. People can start by assigning users identity to agent but it's not, you know, users are trickier than agents because, sorry, agents are trickier than users because code can change easily. So what we done is the following. We want to be the broker who is going to get identity of agents from different parties. Microsoft is very focused on creating agents and their identity. Our exchange can make calls to Microsoft and get the identity verification. Google wants to do its own agents. AWS, Salesforce service. Now we want to get the right identity check things and then make the connection. That is our partnership with Microsoft. We have been working with Entra ID for this agent identification authentication for quite some time. So this is about zero trust agent communication. There's a second part of AI security. In large companies, different users and departments are buying AI application and services and deploying them. So CIOs and CISOs are worried about what's being bought or some of the stuff could leak their data and could cause other issues. So they want to start with AI application discovery and the risk associated with it. Zscaler built applications that are available to our customers to identify what they have. That's point number one. Point number two, they're building their own models and applications either in their data centers or in the public cloud. As they build them, they want to understand vulnerabilities, weaknesses those applications that bad guys can exploit. In the AI world, it's called red teaming, which is a fancy term for pentesting application. We bought Splix because they had one of the most mature applications for red teaming. Then the next thing you deploy those applications. Those applications will be used by some users, maybe your employees, maybe customers, maybe by agents. How do you make sure somebody doesn't do bad things with those applications? What could be the bad things? Well, when you submit a prompt, you can play with things and cause prompt rejections. You could do data poisoning. So, we're sitting in line like a proxy. We are inspecting every prompt. We are looking for prompt for any cyber issues, any data loss issues or acceptable use. What is acceptable use? Somebody could be asking questions to say, 'Give me the list of bad things about the product you're selling.' Okay. I mean, those are not good prompts to be accepted. So, the customers can define a policy. We check the stuff. A prompt is acceptable, let it go, the response comes back. We're inspecting the response to make sure it's a reasonable response. And the results are not really, really bad. We have seen some examples where somebody said how do I commit a suicide? I mean that answer shouldn't be given. Okay, those are the kind of guardrails we provide. So when you can provide full end to end visibility to testing of application and securely running it, it becomes a powerful thing and that's what we offer and that's what sets us apart from others.

Of course. The same thing we did with Zscaler zero trust users today with over 55 million users and doing over half a trillion transactions every day. We check identity either Microsoft for users in the Entra ID or Okta or Ping or SailPoint. We are the Switzerland who wants to take identity from any identity provider and securely connect to any application or service could be on a hyperscaler AWS or Azure, could be a SaaS provider, could be open internet or could be your application or data center. That's the beautiful place we sit in and it's the same thing will carry on with agents as well.

So CIOs and CTOs are finding that IT is playing more and more important and pivotal role. The cloud adoption happened to be able to build and deploy application faster to get value out of it. Being able to analyze those data signals they're creating and now AI is further turbocharging it. They all want to build AI powered applications but they're scared. Even before AI, they worried about ransomware and attacks that are happening left and right. And that's really what gave Zscaler a big push because the traditional firewalls couldn't really help. You know, I sometimes talk to CISOs and they say, 'Assume you are already breached.' I said, 'Man, you're spending so much money on these technologies. What do you mean you're breached? That's a defeatist attitude.' But then they say it's practical because my network is so far-flung there's something breached somewhere and that's how they need to, they're moving to zero trust architecture. That's one piece but AI is making them worried further but in the selling environment there's tightness, there's deal scrutiny and often CIOs and CTOs are asked to reduce the budget sometimes 5% sometimes 10% and sometimes 15%. The cyber budgets are actually pretty stable and have gone up to some degree. When I have conversations with CIOs or CISOs, the two things that get them moving to engage with us. Number one, can you improve my cyber posture? Especially with AI, I'm more concerned even though I was concerned before and that gets them going. But number two, if you can do it without costing a fortune and if we can take a number of point products out and save me a lot of money, then I can get the deal done. In IT, lots of vendors claim ROI and TCO, but very few companies actually save tangible money. Even in the SaaS world, a SaaS provider goes in and said, 'Remove the old stuff. You're going to do customization for 18 months and I'll give you better productivity.' And yes, it delivers productivity, but nothing for 18 months. In the world of Zscaler, you go in and say, 'Mr. customer, you're spending $15 million on these firewalls, VPNs, and other related network security products and you're spending $30 million on this network. And the network is like private roads. Now, internet has become the cyber highway. Most of the network cost should go away.' So, if you do 30 million on network, 15 on firewall, that's $45 million. With Zscaler world I can comfortably go and tell them without knowing their exact environment. Then in about 18 to 24 months I can reduce it by a third to 40%. I mean that's significant savings. That's because I don't have to worry about the cannibalizations of firewalls and routers and switches. Every office simply connects to internet. We are the policy engine. We connect things to things. Lots of these products can be taken out. So when I do this stuff, we create a business case study. Our deals gets done. That's why we're able to deliver some great numbers. You seen us. Our ARR growth was 26% year-over-year. Our revenue growth was 26%. We had a strong billing growth. Cash flow, net cash flow margin is in the 52% range. We beat all metrics from all sides. So, and this is because our customers are getting great benefits from us.

If I may, it's not just negotiating a little smaller. Whatever you're asking me, I'll tell you we can probably save five times more because there's so much old stuff that's sitting in your enterprise.

Yeah. So look, there's a bunch of services customer will have. Some of them have big barrier to entry because of intellectual property. Some will have small barriers to entry. Zero trust exchange for agents I'm talking about is a significant barrier to entry and it'll be hard to have real competition there. Red teaming is a good technology but I don't think over time anyone will have big differentiation in red teaming surface. They'll have similar functionality. Our positioning will be I can do red teaming or any of the application that are being built for on any of the environment. So we become the Switzerland there and some of these other providers they'll always have better stuff for their own models and services and not that optimal for others. So yeah there'll be lots of red teaming providers. Our solution will two things, serving everyone being one and we are essentially making sure it works well for all the parties as a third party neutral company.