Good morning. Hello and welcome to day two of the India Today Conclave 2026. We're in the age of disruption, and the disruption is not just taking place in West Asia and the world in terms of conflict, but more importantly, it's taking place in the world of big tech. And our first subject this morning is designed for the times in which we live: machine versus mankind, agentic AI, and the battle for human agency. I'm delighted to welcome a very special guest. Meredith Whittaker is president of the Signal Foundation, a leading voice on tech accountability, on privacy, and she is here to discuss the issue of agentic AI because as machines grow more autonomous, the question looms: who really is in control? Is the battle between machine capability and human agency? And where should society, each and every one of you here and watching this program, draw the line? Welcome, Meredith, to India. Good to have you here.
Thank you. It's a pleasure to be here. Thank you.
Let's start, Meredith, with that central question, one which should be consuming the world, which is consuming AI at the moment: who really is in control of this crazy world of AI around us? Is there anyone at all, individual or government or organization, that really is in control of this AI revolution?
Well, I hate to break it to you, it's me. I am in control. That's a joke. It didn't land well, but we'll move on. And I think that, just to answer this question, I want to give a little bit of history. And so please bear with me. I have academic tendencies here, but I think it helps in a world of so much hype, where AI is everything and nothing in so many conversations, to ground us materially. So there's a question we might ask: why are all the dominant AI companies right now the same companies that emerged through Web 1.0, the platform economy, to become the hyperscalers and the giants? So we do see a bipolar universe when it comes to AI, where you have the giants in the US and then some in China, you know, duking it out for AI supremacy is the sort of rough narrative. And I think we have to look back to the 1990s and a number of moves that were made in the US to privatize formerly government and university infrastructure, turning the internet over to the private sector in a way that allowed for these platform monopolies. So if you look at some decisions made by the Clinton administration, what they did was effectively hand over these networks to private companies with no restriction. And anyone who knows the history of communication networks understands that they function as natural monopolies. From the post office to telecoms, you always have guardrails around these networks, you know, so that you can manage them, so you don't have that monopolistic tendency. In this case, those guardrails weren't put in place. And so a business model built on scale at all costs around the globe, around data collection so you could create and sell ads, and around infrastructural dominance to support that data, sort of cemented itself among key players in the 1990s. And those are exactly the ingredients that in the early 2010s we all realized in tech, where I was at Google, could be put to good use if we applied them in combination with deep learning, with an old AI approach, you know, approaches from the 1980s. So all of this may sound a little dry, a little academic, but I think it's really important to that question you just asked: who controls this?
So let me just simplify that academic answer to explain the first big challenge. Therefore, to my question, who is in control, is you're just flagging off the fear of concentration of power, of data of millions of people across the world in the hands of a few private companies. Am I broadly correct? Without necessarily having the guardrails that are needed to ensure privacy, safety, and security.
That is absolutely correct. Yes, you distilled that beautifully. And I think we have to look at this. It's not just data in the hands of these companies. It is the models, the AI models that they can train, the infrastructure to train these models. And importantly, it's the distribution networks. It's the cloud services that they can license the AI from, the APIs that people, you know, white label to turn into different products. It is the platforms, the Facebooks, the Instagrams, the Amazon marketplaces they can integrate the AI into for dynamic pricing or for ad serving. And so these resources combined are what's needed to create and then use AI from head to tail. I think what this also gives these entities that we haven't discussed enough is under the banner of intelligence, they begin to have the authority to tell our stories for us, to define us, to define our reality, to define our worlds in ways that don't map to traditional data privacy arguments, but nonetheless need a lot of scrutiny given the power, the authority these models and the companies who control them are gaining as these technologies are, you know, introduced throughout our institutions and infrastructures.
Let me come therefore from what you've just said to two points. Point number one: in India at the moment, AI for the common person, and there are a large number of people consuming it, is associated with ChatGPT, with Gemini, with Claude. You're using it for basic information, but we are already now pivoting into a world of agentic AI, which is the next level, where essentially you've got machines making decisions often on our behalf, and perhaps as a result making life easier for us. And therefore a number of people are going to choose possibly convenience over privacy. What we did possibly when the social media revolution came, we just embraced it, not realizing that our data may be compromised, but we are choosing in many ways convenience over privacy. Do you believe that there is a real threat, or is it exaggerated, that as a result we are eventually going to lead into a world where we're going to surrender human agency to algorithms, that an algorithm is going to set how our lives are going to be led? Do you really see that happening in the next decade, or even in the next year?
Well, it's not an absolute. We're already living that way in some sense, right? If we go into a bank and an agentic system determines we don't get a loan, we have just been told how we live in a certain respect by an algorithm that informed a bank that has very material consequences on our lives. So we already are steered and sorted and ordered and allocated or withdrawn resources due to these systems. I think this agentic turn, however, is very important, and it's very important for Signal. How many people in the room use Signal?
How many people in this room use Signal?
Well, we love to see it. It's...
So that's less than 10%. What's your message to them? What's your message to them?
Everyone else should probably use Signal. I make no money when you use Signal. We're a nonprofit. But we are the gold standard for private communication. Militaries, governments, boardrooms, really anyone who recognizes that private communication, the ability to strategize, the ability to game plan, requires infrastructure that actually keeps your secrets uses Signal. And we've become a core infrastructure for a reason. We are the only large-scale messaging app that is open-source that guarantees the right to private communication. Not because I say so, but because you can scrutinize our encryption protocol. The security and hacker community have spent over a decade looking at our code and have found no active vulnerabilities. And we exist as a nonprofit with one mission: to ensure that this right survives in a world where almost every other service or interaction is collecting data, often letting that data out in a breach, selling it to data brokers, or using it to, you know, model your preferences, train their AI models, whatever else it is.
So to distill what you're saying very quickly, what you're telling those who are not using Signal, let's say for a moment that if you zealously are concerned about your privacy and the privacy of your messaging, you should be using Signal. Am I broadly correct?
You are completely correct. If you have anything serious to say, use Signal. You don't want that coming out in discovery.
But what if, Meredith, the criticism that comes: what if there is an antisocial criminal network? And this criticism has been made in the US and other parts. Use Signal because they will then want to create their own little private universe which is not governed by the rules and norms or laws of the rest of the world. I am an antisocial, a mafia of sorts that is using Signal to pass these encrypted messages.
Well, I think the 'what if there were a bad guy' thought exercise is very convenient for people who want to withdraw rights. And I need to be very blunt about that because we are in a golden age of data collection and surveillance. Law enforcement and governments have never had more access to data. Over the past 30 years, we've become a datafied society in which our interactions, our transactions, our social networks, our employment history, our educational history — I have to take many, many more breaths to get down that list — are in the hands of private and government actors, are now being queried by powerful AI models that are finding patterns and identifying people across those data troves. We have never had more data. The art now is finding that needle in that haystack of data. The idea that the only barrier between law enforcement and catching the criminals is one application that holds the human right — it's a human right for a reason — to private communication sacrosanct and ensures that in a world where almost everything else is surveilled, that is not an evidenced belief. And if you look around at how law enforcement is resourcing their task forces on some of the most heinous crimes, what social supports are going to victims? How are we prosecuting the people who appear in prominent dossiers and files as very clearly at least aware of criminal behavior if those people are very prominent people? How is all of that working on a social and political level? We begin to have to roundly question the assumption that the only problem here is that we ensure the right to private communication. There is no evidence for that.
Which brings me to my second point, because to function well, AI assistants need access to our emails. They need access to calendars, contacts, conversations. Therefore, from what you seem to suggest, we are already building or have already built the most intimate surveillance system that mankind has created. We are already being completely surveyed. Our lives have completely in a way been intruded with. Where does that leave the right to privacy? Do I even have a right to privacy given this kind of surveillance system that has already been created?
That's a great question because with the introduction of these agents, we have to question that, and this is something Signal is very concerned about. So one of the issues we have is that 'agent' is a bit of a marketing term, right? What it indicates is an AI-enabled system. So most agents have, you know, most of the big agents, the Geminis, the Claudes, have a handful of AI models working behind the scenes. But there's software systems that run on operating systems. They run on hardware. They're, you know, and they are increasingly being implemented into operating systems at a very low level. And in order to work, there's two things that are true. An agent to do the things you want it to do needs to have access to data. Needs to know as much as possible about you, or it can't really act on your behalf. And it needs to be able to act. It needs permission to act like it's you. To, you know, it's going to plan a car trip with you and your kids. Well, it has to have your credit card so that it can book the car. It has to have your calendar so it can find a good time to do that. It has to have maybe access to your Signal so it can text your kids and make sure they're waiting at the right time when the car picks them up. Well, all of those things are effectively creating vectors for invasion of privacy. When you look at the security architectures of many of these agentic systems, they are the nightmare of cybersecurity professionals. They are extremely vulnerable. And you know what we have here, I think, is a very uneasy tension in which the amount of capex, the amount of investment that has been allocated to these extremely expensive AI models and infrastructure is pushing a lot of these companies to introduce AI in ways that are bypassing traditional cybersecurity safeguards or safety engineering. And when it comes to integration in our operating systems, where we have three large operating systems we all use — iOS, Mac OS, Windows, Android — this creates a systemic vulnerability that we're very concerned about because, of course, we're an application. We also have to build on the operating system. And if the operating system, as the foundation and the toolbox that we as Signal rely on, you know, cuts a hole between the blood-brain barrier between the operating system and the application, letting people in, you have bypassed our encryption and you have undermined key security guarantees. And so Signal is the example that proves the rule. But this danger is coming for anyone who develops at the application layer and anyone who has to rely on these operating systems, which is all of us. That's your military, that's your mom, that's everyone. And so I think there is a need to press pause and just bring some of the adults into the room, realize there's a lot of hype clouding our vision, and that the consequences for this kind of cybersecurity recklessness are, you know, significant. This is institutions, core infrastructures, you know, highly regulated industries that can be attacked through these vectors.
But how am I going to press the pause button? It's too late to press the pause button. You know that that engine has already rolled on. How am I going to today press the pause button? Do I have the capacity to have my choice? I mean, is it about individual? I've surrendered my individual freedom. Let's be very clear. I've surrendered my individual freedom the moment I've entered into the world of AI or agentic AI. So having surrendered it, now how do I put in safety guardrails to ensure that at least some minimal privacy of my life, of my data, is with me? Is that at all possible?
Well, I think this is a — I love this question. You know, we can look at a little historical antecedent here. How many people here know that the fossil fuel industry invented the concept of the individual carbon footprint? I know that because I read weird stuff. They did that in part because a lot of criticism was coming to the fossil fuel companies for their role institutionally, industrially, and they said, you know, it's much easier to narrate this as individual choice. I feel bad because I didn't recycle. I feel bad. But ultimately, when we look at who could actually shift the needle, it's the large companies who are doing the lion's share. And in a sense, I think with this narrative, we're dealing with something similar. You know, what can I do? Me as Meredith, not actually being in control, can't do that much. But I did just name a fact here. There's, you know, three major companies that control these operating systems. Where does the pause button exist? You know, I worked at Google for over 13 years. We had cybersecurity, you know, security review of every launch. They would look at an architecture like this five years ago and they would say to the person who presented it, 'You're fired because clearly you didn't pass 101. This is not something that we would ever ship to production.' And so the pause buttons exist, but I think we need to, you know, AI is not an ephemeral pollen floating around in our air that just escaped the lab somewhere. AI is a material set of infrastructures built on very specific resources that are in the hands of a handful of actors. And so that does give us a target when we're talking about how do you pause and who actually has the levers for pause. This is not saying no AI. This is saying stop reckless deployments and recognize the collateral consequences of making short-term decisions on the backs of our long-term security and privacy.
You know what I find fascinating at the moment is, for example, with the war on — you've got a situation where Anthropic finds itself in conflict with the Trump administration. The US government wants control over the mass surveillance systems. Anthropic is saying that's not possible. We've got safety guardrails in place. So suddenly Anthropic is unwilling to give the kind of mass surveillance systems that the government of the United States would like access to. So it almost seems as if some of these companies, some of these monopolistic or oligopolistic companies, are actually more powerful now than governments. They are the ones who can actually almost dictate terms to the government. Is that the sort of dark world that we are entering where individuals don't matter, governments don't matter? It is, as I said at the outset, the algorithm engineer sitting in San Francisco or these large companies which are virtually going to dictate terms. Is that the future that we are heading to unless we put these guardrails? Can regulatory systems be put in place by governments? Is it possible for governments to press the pause button at all, or not possible anymore? Too late.
Well, I don't actually think that showdown is decided yet. What we do have is very clear evidence of the type of concentration of power and of its locus that the US state obviously understands these technologies as part of their geopolitical advantage, and they obviously understand the type of dependency on these technologies that they have when they integrate one or another model into their systems. However, these are not technologies that the state could create. It doesn't matter how much money they have. It actually wouldn't be legal for the US government to collect the kind of data that the private industry can collect. This is another Clinton-era legacy. And so this is built on the backs of a business model that has pulled the resources for AI in the private industry that could not be replicated by the government in terms of, you know, world-spanning cloud services, these massive platforms to both distribute and collect more data, and the ability to collect the kind of data that is in the hands of these companies. And so you are seeing a tug of war, a showdown between the imperatives of the state, which is aware, increasingly self-aware of this dependency and its potential danger, and, you know, what we can call the free market and, you know, frankly, the ability to build and shape technologies, you know, per one set of imperatives or another.
You know, just to give an Indian example, in India we're digitizing everything — payments, government identity, even governance. Therefore, in this age of AI, do you believe that empowers our citizens? Governments say that's empowering you. But are we being exposed without realizing it to new forms of surveillance? I am surrendering my individual freedom to the government in a democracy. Is that the way it is? Is that the way you see this entire digital world? Even pre-AI, the way the world was getting digitized.
Well, I think you can't say getting digitized without recognizing that is a form of surveillance. Like the census is a form of surveillance. Is that surveillance used to ensure that everyone has food and everyone has care, or is it used for oppression and division? That's the question. But I think, you know, of course, this is a form of surveillance. You know, data is being collected. It is being, you know, analyzed. It's, you know, that's what it is. I think we need to also recognize that governments are not like sports teams that we root for irrespective of who plays on them. They are containers for power. And so we need to recognize that that surveillance could be used justly and it could also be used in oppressive and harmful ways.
Can I ask a simple question? A lot of people in this audience, somebody's put it: where do you store the data? Where do the Signals and others store our data, and where does the AI model store our data? Where is our data actually? Who has it? It comes back to the question of control. But if you can explain to us, where is this data being stored at the moment?
I love this question because Signal does not store your data. We have built a system in which we have access to as little data as possible.
You don't store my data? Subpoenas come to Signal?
We cannot give over the data because we don't have it. We have built, we have rewritten up and down the stack to ensure that your messages stay between you and who you talk to. Signal, no one else can see those. You're the only ones with the encryption keys. That your profile information, your contact list, whom you're contacting...
