BIG INTV: Signal's Meredith Whittaker Says AI Is Just A Branding Term

0:03From Wired, this is Uncanny Valley, The Big Interview. I'm Katie Drummond. When Meredith Whittaker was an undergraduate studying rhetoric, literature, and psychoanalytic theory at UC Berkeley, she probably wouldn't have predicted her now storied career in tech. But in 2006, the broke recent grad needed a job, so she uploaded her resume onto Monster.com. In between interviews with publishing houses, she was contacted by Google for an entry-level customer support gig. That role evolved into a long and winding upward trajectory before her dramatic departure after 13 years. Maybe she didn't predict all of that, but almost a decade before any of us heard about ChatGPT, Whitaker was predicting something else.

0:44AI was worth taking a closer look at. Her interest, skepticism, and thoughtful perspective on the ethics of AI technology produced friends and detractors, quickly catapulting her into a global thought leader. And while she started her career at Google, the most Silicon Valley of Silicon Valley companies, she's now the first ever president of the Signal Foundation, a nonprofit whose flagship product is an encrypted messaging platform used by millions of people around the world. Among those people, of course, are the Trump administration officials responsible for SignalGate earlier this year, which catapulted the foundation and Whitaker to the very forefront of the political and cultural zeitgeist.

1:26Meredith Whitaker, welcome to The Big Interview. Thank you for being here. I'm happy to be here. Nice to see you, Katie. Nice to see you, too. So brace yourself, because we always start these conversations with a little warm-up. I'm going to ask you some very fast questions. Are you ready? I am. Okay. Mountains or beach? Mountains. What's the most overhyped AI buzzword right now? Agent. Oh, I knew you were going to say that. What's the weirdest AI application you've ever seen? Okay. I'm having to dig through a bag of weird to try to identify something that could top the stack rank. A chatbot that pretends to be your friend.

2:02That is weird. Yes. That's weird, right? Weirder every day. If Signal had a mascot, what would it be? We would never tell you. What emoji best sums up your philosophy on privacy? The ghost emoji. Nice. More secure, handwritten letters or encrypted texts? Handwritten letters. coffee order simple or complicated simple you're she's telling the truth she's drinking what looks like a very basic coffee right now if you weren't working in what we'll call what you do tech if you weren't working in tech what would you be doing what's your alternate career path a poet love that someone asked me that once i don't want to name drop but it was david remnick in a job interview and i said massage therapist oh my and he was like what what is wrong with you You hired?

2:54You're like, David. Yikes. Wow. I can make that joke. It's fine. I'm not in your industry. It's fine. I'm blushing. Okay, so let's talk a little bit about you so that I can stop talking about that very awkward interview I did with David Remnick. Favorite subject. I am so sorry. And interestingly, actually, we don't know a lot about the early life of Meredith, which I realize is on purpose. And you've said that it's on purpose. You've talked about how you've decided to keep your personal life private. You decided that at a very early age, if only more people were so careful. Tell me about that decision.

3:33I don't think it was a conscious decision. It wasn't a flex. It's not like I woke up after reading a book on how to live your life, and I decided this is who I am, and I'm maintaining this firewall. It just seemed weird and creepy. and my friends know me my family knows me but you know I can't I was like a teenager when MySpace and Friendster and these early proto-social networks were coming up I came up in chat rooms and like Usenet groups and it was never really about it was about like is there a persona you could create that's like funny or weird or skewed how good are your clapbacks how accurate is your information can you roast with the roasters like it had very little to do with a sort of expose of your personal life and i think it's strange that that has become part of sort of normal culture that everything about you is assumed to be fair game and then i'm you know like it's a question around like what led you to keep your personal life and your relationships to yourself.

4:49And I'm like, but what led us to assume that every micro action and relationship and social context is mineable and tractable and even understandable from the perspective of data and sensors and giant SQL databases that are mined for typologies that can then be sold to advertisers who want to reach a certain type of person? I don't know. There's nothing complicated about it. I think there's something weird about the time we live in and the way that every single human action and relationship is assumed to be scrutable and mindable and, you know, share game, basically. Yeah. Okay. Well, look, fair.

5:33Fair answer. We do know. We do know about you. Who do we? My producers and myself. We do know you went to UC Berkeley. I did. Right? You studied rhetoric, English literature. What did you think you wanted to do? I thought college was a much easier hustle than working retail and managing bands and scraping together money. Like, it wasn't—I didn't come from a place where you sit down a five-year plan as a 12-year-old and, like, get into school. It was sort of, you know, it was hustling. It's hard to work retail. At first, it was just a much easier job. And I was—I'm good at school. I was always a big reader.

6:11Like, that was never hard for me. But it wasn't a career aspiration. It was like, I'm really, I love reading books. This is the most fun I could have, you know, working, in quotes, right? And then I graduated. There was some professors who wanted to sort of push me to grad school. And I was like, that is not a good bet. I don't want to take out any loans. Like, I had this sort of class-based caginess around owing money. And then I started looking for jobs. And I put my resume up on Monster, which is old LinkedIn for the youth. Oh, yeah. And Google recruiter reached out. And so I was about to go there.

6:49Yeah. And we're not going to spend a ton of time on Google. But speaking of fun and fun things to do, you worked at Google. You worked at Google for some time. Yeah. Was that a, like, culture shock feels like even the wrong way to describe it. But you at Google, I will say, like, knowing you now is hard to imagine. Did it feel hard to imagine to you when you were sitting in your cubicle, drinking the free juice, eating the free lunches? It was odd. But the jump from where I came from to the lowest ranks of Google when I started is farther than any jump I could make from my sort of entry-level position at Google to anywhere.

7:28It was just a very, very different context. And so I didn't know how abnormal that place was. I was just like, I guess this is what a business job is like. Right. All right. We sit a lot. You do not have a social life. You get up at six, you take the shuttle from Berkeley to Mountain View, and then you get home around nine and you sit a lot. But also it's like a feral teaming environment. And at that point, when I started in 2006, it was a very different tech context. Yahoo was a larger enterprise as measured by monthly active users. The IPO had happened a couple of years before. So Google was, there was an exuberance there.

8:12There was a huge amount of money, but Google had not entrenched itself as a dominant monopoly player, right? These were sort of platforms trying to kind of monetize attention in various ways. And there was a real belief at the core of that culture that finally the formula to being ethical billionaires had been found. We'd unlocked, you know, the good kind of capitalism and this was it and the world would be transformed. And, of course, on the backs of Stephen Levy and these sort of, you know, utopian dreams. And I was like, whoa, this is interesting. There was also because I think there's a lot of money and because at the time I entered tech, the people working in tech along with me hadn't gone into that field largely because it was a money field.

8:56They'd gone into it because they were like wooly little nerds who loved circuits. And that's not the case now, right? Yeah. Like if you went into the money field, you were like studying for the bar and the LSAT. You were studying for the MCATs. You were like doctor, lawyer. those were where the kids who just had sort of steely ambition or like really hard-edged parents were going. But there was a sort of genuineness to a lot of the people. There was a huge amount of tolerance for like weirdos and nerds and kind of like strange, odd people, which I love. That's my crew. And so in that way, it was a pretty fun environment because if you worked at Bell Labs, you were hired.

9:36If you were working on some core protocol, you were hired. And so a lot of very smart people who were genuinely interested in what they did were roaming around with money and time and a lot of permission to do stuff. And I, the lucky part about coming where I came from is I didn't realize those checks weren't written to me. So I was walking around trying to catch them everywhere. I was like, well, I want to do that. And they were like, well, you know what? We didn't hire you particularly to do that. You did not invent a corporate, but you're in the kitchen, you're cooking. Okay. No one likes conflict in this environment.

10:06So I guess we'll let you proceed yeah yeah yeah it's so funny to think about that now when you know you're you're going to work there to wait out your four years so that your options vest and then you you're on to the next it's a it's a money machine yeah I mean you know at some point they brought in McKinsey at some point like you can't just continue reporting growth and revenue on the backs of pure ethical. You know, how we've described it in the past is the, at that point, the horizons that, you know, of like needing to do bad or leave money on the table were very, very, very far in the distance.

10:53And ultimately, the machinery of the publicly traded corporation doesn't tolerate that type of bright line. And so I think that has a lot of explanatory power for like, Why from that to this? One of the interesting things about your time at Google, to me, in thinking about where you are now and where we are now, is that you watch the evolution of AI, right? That was unfolding before your eyes. And I think before many, many, many people in the mainstream were like, oh, chat GPT, generative AI is a thing, which is really the last two or three years. But you have actually been warning about the implications of AI for many, many years.

11:32So you did an interview in 2019. I think it was with Kara Swisher. that's six years ago now, where you talked a lot about this. You described sort of the crap in, crap out process. Can you talk maybe a little bit about those earlier years looking at the evolution of AI and looking out into the future and seeing things that we are now experiencing or even then were experiencing, but sort of weren't in the mainstream consciousness? I mean, did you feel crazy? Were you looked at as crazy to be sounding those alarms early on? Yeah. I mean, I never felt crazy. The closest, I think, was like, oh, I must not be clear if this isn't landing.

12:15Like, how do I say this more clearly? Because obviously, if people heard it, we would act, right? And that's a more naive version of myself. But, you know, I started 2006 at Google and then I built a co-founded an effort called M-Lab that was large scale network measurement. And that's network measurement is like basically measuring your Internet performance. Like, why isn't Comcast working? And this was an effort to create data that could live in the public domain that could allow us to understand net neutrality, which was a big topic back then. And also just to have a public source of information around like how these core infrastructures, these telecommunication networks are working.

12:57And in that process, I got really sensitized to the process of creating data, right? Like the millions of kind of editorial choices that go into, you know, how are we designing a methodology that will create data that we can be comfortable saying is a proxy for a very complex reality. I was already backed in to just how contingent and like ultimately editorialized data is. It's not a flat reflection of the facts of our universe. it's a series of choices that people have made that reflect their positions, their interests, the job they have, right? Why was Google paying me to do this? Well, it had an interest in this data set existing, right?

13:42They were paying$40 million a year just in bandwidth to help sustain this infrastructure around the world. Why would they pay that? Because that data set was doing something for them in the world. And so that's happening during the 2000s into the early 2010s. And I knew machine learning was around. Machine learning was one of a number of statistical techniques that usually lived between ads and research, help optimize advertising auctions and ad placement, whatever. But it wasn't the godhead, right? It was like a thing some people did. And then it was early to mid 2010s that I started seeing like machine learning courses pop up.

14:25It became a thing that would move from a little tech talk around a machine learning innovation to the main stage of Google's all company weekly meeting. But there was this meeting I had at that time. I was managing a pretty big budget. I was running the open research group, which was collaborating with a lot of academics and open source projects around big issues that span beyond Google in tech. And it was a lot of fun. And I was funding a number of projects so people would kind of come in and pitch me. And there was this team from Harvard that came in wanting me to fund an effort to create a kind of machine learning based genocide detection algorithm.

15:08Wow. Okay. And, you know, this is, I don't know, 2015, 2014, something around that, 2014. And I was like, wait, what? You know, how do you define that? Is that Lemkin's definition? Like, how do you augur that with data? And if you look back at the We Charge Genocide petition that accompanied sort of the U.N. adopting Lemkin's definition of genocide post-World War II, there's always been contestation around who gets to claim that term and who doesn't get to claim that term from the very beginning. but I'm asking these questions and I'm getting no answers. It's just a sort of tag back till the math will solve it.

15:46We'll build a model. We think we can get the data. And it's like, well, census data across countries isn't even fungible, right? It's collected in different ways. It represents totally different methodologies. Like, how would you do this? But that to me was this moment where I was like, okay, what are we actually doing here? Because I'm seeing this sort of pop up around the company. There is this claiming to be able to predict, detect, do things that I know the data they would be using is way shoddier than the MLab data. Like, I know how obsessive we are about methodological rigor, about all this, for measurements that are very low level, arguably some of the most objective, in quotes, measurements that you could do.

16:28And still, they're fuzzy and we're constantly having to make editorial choices. Now you're saying you can model this with social data, with census data, with data that is like in no way as rigorous or as easily objectifiable, to misuse that term. And that was the catalyst for me beginning to get interested in that field, read a lot, and kind of recognize that what was happening was some sleight of hand and actually had some pretty significant, I think, social risks to it. And so then fast forward to now, the year 2025, I can't get through half an hour in my job without hearing about artificial intelligence.

17:10It has taken over the world from optically, optically, not in practice, but optically. What is your assessment of the state of play with that industry? I mean, you've written before, and I think you wrote a piece for us, actually, I think at the end of 2024, where you talked a little bit in that piece about sort of the AI bubble on the precipice, right? That there were signs that this was starting to crack, that we were starting to see, I'm now mixing metaphors, but this bubble bursting, cracking, bursting, blowing up, whatever. Where are we now on that journey? On that journey. Yeah. That piece I felt like was, I remember asking the person who was editing it, I was like, I don't want to do a prediction.

17:58Can I do a manifestation? Yeah. So I think I have sort of like two answers to that. One, like AI is not a technical term of art, and that's very convenient for this market, right? It was a term invented in 1956 by John McCarthy, who was a cognitive computer scientist, and he wanted grant money, and he wanted to be the father of his own field. Right. So that term was invented, but it was actually not referring to the neural network paradigm we're now using that now is what we mean when we refer to AI. It was referring to these symbolic systems, which at that point were opposed to the neural network approach, the connectivist approach, which predated the term artificial intelligence by over a decade.

18:47So neural networks, it was like 43, I think. You had sort of the term artificial intelligence invented around 56. It didn't apply at that point to neural networks. It's kind of this blanket that has been thrown over a rotating set of technical approaches that currently, in this current moment, refers to these very, very, very, very, very large-scale models that require huge amounts of compute and huge amounts of data and certainly do really, really impressive things like trick me into thinking they love me or whatever. And this blanket could arguably be thrown over any other paradigm, right, if we decide that AI is sort of describing something else.

19:28That's a long way into saying, like, I do think there is a, you know, there is a threshold at which this, you know, the metastatic logic of the current AI moment, you know, bigger, bigger, bigger, bigger, bigger, all the water, all the energy, all the data, you know, all the every compute is going to give out. You have kind of climate thresholds. You have just thresholds on how many data centers you can build. Does that mean there's actually sort of a faltering in the AI market? I don't know. And I'll just end that by saying I think one of the reasons we're seeing this kind of glassy-eyed embrace of so-called AI agents is that they're really trying to find this consumer market fit for this bigger is better paradigm.

20:26And, you know, in a sense, that is just a sort of new branding. It's like wrapping everything we used to call assistants in, in like a new brand term and kind of hoping we can make that work. But there is a point at which the trillions of dollars that are being poured into AI, you know, either will or will not find that killer application and begin to break even. And we have not reached that point. And so I think that there is a threshold coming where the market will rein in the leash. And I don't actually know what happens then. Do we just shift what's under the hood of the definition? Move the goalposts.

21:07Yeah, yeah. To move the goalposts. Look over there. Now, I want to talk to you about signal, which seems like a relevant thing to discuss with you. I love talking about signal. And I'm going to start by, I'm just going to, good. I'm going to start by bringing AI into it because I remember we had coffee six months ago. I remember you talking to me a little bit about the risks of this moment in AI for Signal, particularly with agentic AI. And I'm curious about, given the moment we're in, where it leads us TBD, but given the moment that we're in, what does all of this mean for Signal, actually? Well, the first answer to that is we are seeing steady growth.

21:46I think we are in a moment where the stakes around privacy and private data are coming home to people in a personal, emotional, nervous system level. It's not an abstract concern that you've got to get your laundry done and then maybe you can think about that. It's data breach, data breach, salt typhoon, geopolitical fracturing, what have you. People are sensitized, right? Oh, absolutely. So in that way, I would say it's a good moment for Signal, and we are very proud that we are there and ready, and anyone who needs us can download it and use it. This is not a prefigurative, speculative project.

22:28This is infrastructure that is in place and ready to meet the moment. Very grateful to the team and the founders and everyone who made that possible. And then, on the other hand, there are real dangers to Signal in this environment, particularly, you know, there are, I would say, misguided or malignant legislation that is continuously proposed that would ultimately make it impossible for Signal to provide the robust privacy guarantees that are our entire thing. You know, undermining encryption, giving governments access, you know, things that just completely nullify the point of Signal's end-to-end encryption.

23:09And there are AI agents that do things for you autonomously. And in order to do that, need access to all of your data, your apps, et cetera, in a way that is pretty unprecedented. And the issue for Signal is these agents are increasingly being integrated into the operating system in ways that the operating system that Signal and all applications need to run on, whether it's iOS or Android or Windows. And in order to work, they need access to your data. So you have these systems that are, you know, offering to, like, book a restaurant for your friends and, you know, tell your friends the restaurant is booked and then you put it on all your calendars, right?

23:55Yeah. And in order to do that, they need access to your browser, search for a restaurant, say, access to your calendar. They can see everything else that's going on there, who you're talking to, access to your credit card to put the down payment for the restaurant, access to your, let's say, signal in order to... Message your friends, access your group chats, text your friends on your behalf. Pull down the information your friends reply with, like, hey, yeah, I'm free on Tuesday. Or is it your friends or their agents? You don't know. And then, you know, act on that without asking your permission.

24:29And almost certainly sending that data off of your device to be processed on cloud servers because there's no on-device models. Right. So ultimately what we're talking about is a backdoor. And we've already seen this, you know, in the context of WhatsApp. There was a report from Lumia, which is a security research forum, that showed that, you know, WhatsApp messages were being sent to Siri. Wow. And back as part of the latest Apple intelligence rollout. To me, that is nullifying WhatsApp's end-to-end encryption promises and, you know, ultimately creating, you know, the backdoor is now that vector from WhatsApp to the Siri servers and back.

25:12And so given that environment, how does Signal ensure that what happened with WhatsApp doesn't happen with Signal? You know what I mean? Like, how do you protect your users? Well, according to the Lumia report, it looks like WhatsApp worked with Apple on this instrumentation. I don't actually have the details there. Signal is doing whatever we can. To be clear, the future of total infiltration and privacy nullification via agents on the operating system is not here yet. But that is what is being pushed by these companies without the ability for developers to opt out. So we have spoken out very strongly about the existential threat that this provides to signal and application layer privacy generally.

26:01And there was one kind of concrete example where Microsoft rolled out a agentic system called Recall, which is the conceit there is that instead of you having to remember what you were doing on your browser three and a half weeks ago, Recall will tell you. So you're like, hey, you know, at 3 a.m., what was I doing on my machine? And Recall is like, you know, you were doom scrolling. And you're like, oh, thank you, Recall, for reminding me of that precious moment. But like how does Recall do that? It does it by screenshotting your desktop every few seconds and then running those screenshots through an ocular character recognition system, which basically pulls out the text and then stores all of that in a database.

26:48And the first designs for recall were to store that in an unencrypted database on your desktop. Of course, that would include your Signal desktop messages. So again, you've just created a vector there. Total nightmare situation. A total nightmare situation. We should not be having to make those tradeoffs. And so what we're calling for is, you know, very clear developer-level opt-outs to say, do not fucking touch us if you're an agent. Signal is entrusted with some of the most high-stakes communications in the world, like real life or death communications. Militaries, governments, every boardroom, human rights workers, dissidents.

27:27Ultimately, if you have high-stakes communication, you are almost certainly on Signal. That includes many of the executives of these companies. That includes many of the governments that they work with. And so it's, you know, this is not a, we believe there is a, at least a window of interest convergence here where this is not a gamble we should be making. And this is something that even if the rush for breakeven, for finding that market fit that has been elusive so far is, you know, pushing down from the boardroom to the C-suite to the product managers, like hold up the collateral consequences of this are pretty dire.

28:07In 2021, and I hate to read your own writing back to you, but I will. You co-authored a piece in The Nation where you warned at length about the risks of big tech being leveraged by authoritarian forces. And there were a lot of lines that stood out to me, but I'm going to read this one to you. The neoliberal bargain is fraying. And if we don't vie for control over the algorithms, data and infrastructure that are shaping our lives, we face a grim future. It is time to rally behind a militant strategy that recognizes the danger of leaving U.S. tech capitalists at the helm of systems of social control while far-right authoritarians jockey for access.