Good morning. I want to thank Pari and the Zinnov team for putting on a great, great conference and for the invitation. Thank you. Thank you. So, I was backstage chatting with folks back there and apparently it's rude not to acknowledge non-human identities in the room. So all agents welcome also, you know, we got to make sure too early. Okay. I'm going to spend a few minutes hopefully sharing a little bit about a new discipline: how you think about resilience in the era of AI. We're all excited about how it's changing the world. It's changing enterprises. It's changing our businesses. It's helping us reshape how we do things, rethink what productivity looks like. It's democratizing our ability to deliver. Growth is thought of differently. How fast can we do things? Costs are looked at differently. We had cost per head, now we're thinking cost per token. It's changing everything. But it's also very exciting. It's moving markets. You look at the stock markets, AI is moving it. And it forces us to rethink how we evolve ourselves as human beings, as contributors, as professionals, and what we do and how we do it. So, it's super exciting. And I've been in this industry for three decades, and I'd say to you, there's nothing been as exciting and as real as you can get your hands on as with this technology. So I'm going to spend a few minutes this morning sharing what I call the AI resilience reckoning. Okay, so let's dive into it. Ready? Okay.
Every technology wave, every technology wave, every revolution if you would, has some kind of a reckoning. So those of us who've lived through the early days of the internet, it was all about privacy: What were we sharing? Where was it going? Cloud was all about security. I'm oversimplifying it, but in principle that was the reckoning. And with AI, it's resilience. You may not have thought of it that way, but hopefully we can have a discussion around it. So, what you want to think about is that in the enterprise—I'm talking about businesses that you and I are part of—in the enterprise, the reckoning is the moment when you discover that the true limit to AI autonomy is not what it can do, but its resilience. How solid is it? How much can you trust it? How much can you rely on it? What's the stability of it? Can you put your reputation as a business to say we trust it? So resilience becomes really important as your business becomes more dependent on all things AI.
So we'll talk about a few things that I believe will be part of this reckoning. Hopefully we'll be on the same page. The first is the AI adoption paradox. While AI adoption is moving at an unprecedented pace, how many of you ask the question, what's the real value that we've been able to get out of it? Right? Your bosses ask you that. My board asks me that: What's the value? How do you measure value? And you tend to lean on what I call vanity metrics: "Oh, we do this faster, we do this better, we do this." But at the end of it, how are your businesses truly growing? How are you getting value? How are you as an individual, as a manager, as a leader contributing to that next wave of growth of your business? So the paradox is there is a lot of it coming into the enterprise rapidly in all forms, but we all have to sit down and figure out how the value really is being shown, is being measured. And for most companies—and I talk to customers around the world—we're at different stages. Everyone wants to adopt it, everyone is trying to adopt it, but you're at different stages of embracing AI in the enterprise. This is a simplified way of thinking about the journey to becoming a frontier AI company, a company that leads with AI-based solutions or capabilities. And if you look at this little chart, it's not perfect science, but it tells you over the last three years how much has evolved in the way we think about AI. It went from chatbots to copilots where humans were directing and AI was doing some of the work, and then agentically where AI is executing, and then learning loops where things get more intense and AI does more AI and agents do more agents. So in a short window of time, the enterprise has had to deal with this in a rapid way and trying to embrace it in a way that's not only productive, but also safe.
So the good news is that enterprises in India are ahead, on average, than companies worldwide on adopting AI. All the conversations this morning are about how deep AI is in everything we do here. Whether you're a GCC or you're an enterprise building a business, AI is an implicit part of your thinking, of your value proposition, of your future: 40% compared to 28%. This is a recent study by Deloitte. Plus, minus, the good news is you're not behind. You're embracing it and hopefully creating things that add value with AI. Now, what's important to understand is the velocity of technical change that's coming our way. I showed you the chart just from copilots to learning loops. And this is where our company really thinks about it a lot. The science problems we like to solve for is the pace, the velocity, the volume, the variety of data that AI uses, consumes, and creates. Think about that for a second: data. So with the velocity of technological change, what happens here is data has always grown regardless of the economy, regardless of the stock market, regardless of the country. Data has been growing always. It's never been at the point where you say, "Oh, data has slowed down." Data keeps growing. And a recent study by IDC estimated that in the next three years, by 2029, we'll be producing about 17 petabytes of data a second. A second—whether it's 15 or 20, it's still a lot per second. Now, that data will live across your organizations. That data will be in flight between systems. That data will go further away from where it was ever intended to be used from a system of record, learning, teaching a system, training an LLM, doing transactions. That data is going to be everywhere. And that is what needs to be managed. That is what needs to be secured. That is what needs to be governed. And that is what needs to be recoverable because things will go sideways. By definition, a learning model is exactly that: it learns, and sometimes it learns right, and sometimes it learns wrong. Sometimes it goes sideways. Sometimes bad things happen. What is your level of confidence, trust that this technological change and these massive amounts of data that have been created are actually being secured, governed, managed, and recoverable by your business? And that's the crux of what we think about. And that's why the reckoning for AI resilience is you can't think about this post facto. You have to start thinking about it active-active. As you create, you make sure you're resilient. As you create, you make sure you're resilient. Because if you're not, what used to be an hour of tolerance yesterday—"Oh, my system went down or something bad happened or a malfunction happened, I've got an hour to recover it," that was okay maybe—now, in the AI era, could cause catastrophic damage if your business depends on that system. I'm not trying to be dramatic; I'm just trying to be factual. Time spans have changed. Your expectations from what your AI can do for you have changed rapidly in the last 12 months, let alone the last four years. So think about it that way, and the pace of change is really the reckoning.
So if you think about this, AI is available everywhere. It's been democratized. Everyone has access to it: the tooling, the ability to really do things with it. And I came across a great blog written by Satya that really talks about the priority for all of us as we use AI, as we embrace AI, is around building a frontier ecosystem—a community, an ecosystem that has the learning loop where your organization is in the middle of it all. And you encode the institutional learning and knowledge, and that compounds both human capital and token capital, AI capital. And the speed and scale of this technology is, frankly, sometimes outpacing our ability as humans not only to adapt to it, but also to be able to use it. Now, it's mind-boggling sometimes, and being in the tech space, when you see the things that are happening and the kind of things that can be done with AI, I always come back to: Is it resilient? Is it recoverable? That's how I'm conditioned now. I'm sure all of you saw a few weeks ago that Anthropic, OpenAI, and Meta disclosed that their AI systems went a little out of the box and were targeting other companies in episodes. Now, this is life in some form imitating art. It feels like a movie. How many of you read the Hugging Face article and said, "This can't be real. This feels like a movie. This feels made up"? Right? You read it, you go, "Wait, it got out of the sandbox, and it went out to Hugging Face, and it broke into Hugging Face, and it took what it wanted, and it learned what it had to, and it came back." And then you think about it, you go, "Wait." You saw that video at the start of the day where you said, "Is it going to get smarter than humans?" The fact of the matter is it's incredibly powerful. The fact of the matter is that it's going to keep getting smarter, and our ability to play a role in that is the reckoning that we all have to deal with. And resilience becomes key because when something goes sideways, you need to be confident that you can bring your business back to exactly the state it was, without the shadow of a doubt, before it went wrong. That's resilience. That's the reckoning.
Which brings me to the fourth factor in all of this: zero-day threats that AI can enable. We're at the beginning of this, if you would. We're not in the throes of it yet, but the potential of what can happen is very visible to all of us, which is why I always talk about active-active. As you build systems, you actively protect those systems. You can't wait to do it later. You have to build that into your engineering mindset and process up front. So, I was a CIO many years ago. I experienced a nation-state attack as a CIO. Threat factors have changed, but the implications of an attack haven't changed. What has happened though with AI is there's incredible compression between the time an attack starts and an attack can be successful. And our job is to obviously make sure that that doesn't happen. But if it does, are you resilient enough to come back? Now, there's a company here in Bangalore, CloudSEK, that we work with closely, that is doing some incredible work in finding out what AI-based systems, for example, are vulnerable and what the implications are. They recently put out a study which showed that already the largest AI supply chain attack breach this year—there's a company local here and they published this report worldwide—over 2,000 companies were affected and there were over 400,000 CI/CD pipelines that were corrupted. And this is targeting AI companies, this breach. So, one poisoned update implicated over 100 AI vendors and over 2,000 downstream organizations. And we're just at the start of what could be the downside of some of this powerful technology.
So what's important here to understand is—and I love that quote—we spent 50 years securing what computers store, and we'll spend the next 50 securing AI infrastructure. Well, it probably won't take 50, but we have to secure the infrastructure that we're investing in and building everything around. And sometimes that's an afterthought, and our premise is it needs to be actively as you build it. That is the big change: active-active. There was another study by Akamai that said that 17%—now this number is inverse—17% of Indian organizations felt that they were well-prepared to manage an AI-based security risk. 17% versus 40% of companies that are embracing AI, and 42% say sophisticated attack techniques are their primary cybersecurity challenge. The blast radius of an AI attack is what you have to manage, and then being able to recover with clear provenance of what happened is what we all have to think about.
Now, it's going to become, in my opinion, the survival of the most resilient. And it requires new approaches to security, identity, and your ability to recover confidently. So, as we roll out these systems, we had that thing that Pari put up in the morning that talked about the negative and the positive. And this is not about being negative or being positive; this is about being prepared. This is about really making sure that you are thinking through the implications of what we're building with all the upside that AI brings us. So it's the survival, in my opinion, of the most resilient: security, identity, recovery. I want you to pause and think about these three things. You have to secure, you have to put policy, you have to put process, you have to automate your data security. You can't go back and do it later because there's more data being used and more data being created than ever before. And AI loves data. It lives on data. The lifeblood of AI is data. And there's a lot of data that needs to be secured. Who has access to what? Where? Who touched it? Which agent? Which human? And if agents are going to outnumber humans 30, 40, 50,000 to one, identity becomes paramount. What identity did that? Non-human. We all know how to manage human identities, but non-human identities proliferate. They're powerful. They have privileges. They have access to data. We give them the authority to do things on our behalf. They can call other agents. Those agents run in long learning loops. How prepared are you? How resilient are you to come back if something goes sideways? So identity becomes really important, and not just identity management—identity resilience. Can you say what happened, by who, which identity caused the problem, and what the ripple effect of that was all the way down, so that you can bring it back without the shadow of a doubt in seconds, in minutes? Because hours are maybe too long. As AI becomes mission-critical, this becomes mission-critical. This is active-active. And then recovery: How bulletproof is your ability to recover? AI systems have tentacles. They touch many things. Agents do different things. Agents talk to different systems. You have traditional data stores, you have cloud-based data stores, you have external data stores, you have lakes, you have inference, you have knowledge, you have skills. All of that composes your system. How confident are you that you can recover that with the click of a button to the state it should have been before something went sideways? So security of your data, identity as to what human or non-human identity is touching and causing something to happen, action to be taken, and then your ability to recover. This becomes paramount.
With every great technology shift, you have to have a procedural approach to it, a methodology. When we did DevOps, it changed everything from the other ways of building applications. The way we managed security, the way we managed access in the cloud was very different than we did on premise. The way we measured costs were very different, and so you had to change. But with the speed of technological innovation around AI, we believe that we have to simplify this. You have to think about a methodology or a capability, a discipline, an operating discipline around AI that allows you to really be resilient while being operationally efficient or excellent. And we termed this ResOps: resilience operations. How do you operationalize resilience into your AI infrastructure, into your AI application development, into your management as you build it? So very simply put, it's a continuous loop that you have to think about. This isn't a product; this is a thought process, a framework, a methodology, an approach where you discover and classify your data actively as it's being created. You can't do it post facto; you have to do it actively. As you discover it and you classify it, you're protecting it. It's just automatic. You have policy that does that. You continuously detect any cyber attack because time is of essence. Once you do that, and if you find an anomaly that needs to be dealt with, you can automatically, confidently, and intelligently sort of recover it and put your system back—trusted data, clean data, secure data back into production—as if nothing happened. This is the operating thought process that you have to start implementing as you roll out, as you build AI capabilities in your organizations for mission-critical enterprise applications. This is how we think of it as your AI reckoning. Active-active: actively build, actively be resilient. Very important. This is a framework under which to think about it. I put this slide back up because I think it actually is very powerful to say that the limit is not what AI can do for you; the limit is your resilience around that, your ability to confidently be bold, to build things, to try things, to implement things, knowing that if something goes sideways, as this is new powerful technology, that you can confidently hit a button and come back to where it needs to be. That is the AI reckoning. And that's where we are. So with that, I'd like to say thank you for your attention. Hopefully this was helpful. If you remember nothing else, think active-active: you build actively, you protect actively. And that gives you the confidence to really harness the power that we all have in democratized frontier AI. Thank you all.
