Charlie Lin
When an AI agent causes harm, who pays?
In this interview, Charlie Lin, Senior Director and Co-Lead of the AI Foundry at Gen, explains why consumers are the next wave of AI and why security, trust, and accountability have to come with it. He describes how Gen's Agent Trust Hub and Skill Scanner responded to malicious third-party skills during the OpenClaw boom, why he expects a major consumer AI attack, and the unresolved question of who insures an AI agent's actions. He also explains how the Foundry runs like a portfolio of startup bets, what Palantir's early ontology investment taught him about preparing for technological change, why he thinks hardware could unlock consumer AI, and why the wave of AI-driven layoffs has gone too far.
Charlie Lin is Senior Director and Co-Lead of the AI Foundry at Gen, the consumer cyber safety company behind Norton, Avast, LifeLock, and MoneyLion, where his team builds new AI products for consumers.
Before Gen, Charlie led enterprise go-to-market as the first business hire at Codex, a stablecoin infrastructure startup. He spent three years at Palantir, first as a forward deployed engineer and then as a commercial tech lead for AI deployments. Earlier, he worked on deep learning for Google Ads and interned at Bain and Blackstone.
“The next big thing is going to be triggered by something huge in the consumer space, where one big consumer hack affects the lives of a ton of people.”
Charlie's core argument is that consumers are the next wave of AI, and that the agents acting on their behalf need security, trust, and accountability built in. A major consumer AI attack is coming, he believes, and nobody has yet worked out who is responsible when an agent causes harm.
"No one today can really insure the outcome or the action taken by any specific agent," Charlie explains. "At what point in the value chain does the insuring start or stop?"
The conversation
In this conversation: Josh Rubin (Host, CTO Studio) and Charlie Lin (Senior Director and Co-Lead, AI Foundry, Gen).
I'm the senior director and co-lead at the AI Foundry at Gen Digital.
And what does the AI director at this Foundry at Gen Digital do?
Yeah, maybe I'll start by telling you a bit about Gen. Gen is a global public company dedicated to serving consumers and giving them digital freedom, specifically through our portfolio of brands, including Norton, Avast, LifeLock, and MoneyLion. You may have heard of these. Traditionally, we've really excelled at providing cybersecurity and cyber safety, as well as financial wellness, for our consumers. We think AI is touching every business, every government, and even consumers, and consumers really are the next wave that has been untapped. Most AI revenue today is in the enterprise, and how much value it can add to a specific business. The next wave is consumers starting to use AI. Today, the most prevalent consumer AI tools are things like ChatGPT, and people use it almost as if it's Google: to answer questions and solve basic things for them. But there's going to be a next wave where more deeply intertwined agentic experiences enter consumers' lives, and we really want to be a part of that. We've been securing consumers' lives for a bit more than three decades, and we'd love to be part of the new revolution as well. Whether it's developing agentic applications directly for the consumer, being side by side with the agentic or model-layer providers, providing security, trust, and peace of mind for our consumers, or even doing very bespoke things, like providing new frameworks for AI companies to know what consumers want and how to best serve them.
Or protecting my father from deepfake scams that are coming at him. When you're talking about Norton, Avast, LifeLock, I would imagine a large portion of your customer base is, excuse me, older. They were the ones with money as the internet was coming up. They're working with an old PC, Norton was packaged in with it, they got LifeLock because they got worried about something. Are you expecting the Boomers as a generation, or even elder Gen X, to start using AI? And how are you protecting them against a wave of threats we have not seen in a generation?
Yeah, that's a great prompt. I think some of the biggest AI headline news this year has been around cybersecurity. If you look at the Mythos model launch and the controversy it caused for Anthropic and the US government around the time of its release, cybersecurity, and the new attack surfaces and therefore the new defense surfaces that AI has opened up, is very top of mind for all companies. The first thing people often think about is government and the enterprise, but consumers are definitely next. We anticipate, at Gen and more broadly, that there will be a huge consumer attack that really brings AI safety, AI best practices, and closing loopholes to the consumer forefront. We don't think that event has happened yet, but we're actively launching new products and thinking through the best ways to do this. On your prompt about older consumers, AI adoption amongst older consumers is less widespread, obviously. I think part of the challenge with AI is that the model makers and the kingmakers of the chips, the Nvidias of the world, are kind of the kings and dictators of this world, and we're just playing chess and trying to catch up with them. It's really hard to say, and this is the challenge all AI companies that are not at the frontier level are facing today: when will something happen in AI? The when is really important, because if you're building something, you're either too early or too late. You kind of miss the wave and it goes past you. So we spend a lot of time thinking through the macro: "Hey, when is something like this going to happen? How can we position ourselves to be a few steps ahead of it and make sure we bring the right product to our consumers?"
Well, if the frontier models exist in the middle, you can either go upstream of that, into the infrastructure layer, power, hardware, et cetera, or downstream, into the application layer. You're very firmly planted, I imagine, in the application layer. So it's less about keeping up with the frontier models or the Nvidias of the world. I don't want to give my father access to my Anthropic or Claude account. I don't know what he would do with it. It wouldn't be good. And in some cases he needs to be protected from it. I've been talking to CISOs about this all the time. Detection is easy. The signals are there, but it's the pattern recognition. There are so many signals coming at us that our cognitive load is consistently overwhelmed by the number of attack vectors. So AI is the only tool that can be used to defend against the problem AI has created. For a company like yours, doing R&D in a space where you have vast experience in consumer-grade protections, whether that's fintech protections or antivirus or phishing, where are you thinking is the best place to spend your time to do the maximum good?
So again, I think it all comes down to timing. Being too early is not good, and missing the wave altogether is also terrible. An example of where we capitalized on this AI security wave: earlier this March, we launched something called the Agent Trust Hub. This was around the time OpenClaw was blowing up the world, if you recall. Part of what made OpenClaw so powerful was the ability to use all these third-party skills people have built to give the agent a harness, to dictate tasks within specific parameters so the output is more controlled and not just a non-deterministic, free-flowing LLM. Around that time, tons of people were using OpenClaw to install numerous skills. In fact, people would just say, "Hey, go do this. Don't even warn me, just go install whatever." And there were a ton of malicious skills being put out into the system that were causing issues for consumers and businesses alike. So we launched the Agent Trust Hub, and specifically a Skill Scanner within it. We did a partnership with Vercel to distribute it, and to date the Skill Scanner has scanned tens of millions of skills. That helps people have more peace of mind, whether it's the agent deciding which skills are installed or the end consumer, that whatever skill we've installed is safe, vetted, with no vulnerabilities. So that's how we captured that wave right as it got hot. Now we're contemplating what the next big thing is. I think the next big thing is going to be triggered by something huge in the consumer space, something that blows up CNN and all the newsrooms around the world, where one big consumer hack affects the lives of a ton of people. We want to be one step ahead of that. So we have very active threat lab teams, a lot of thought leaders in AI thinking about what's next, and we're constantly talking to the labs and seeing what their movement is as well.
That speaks to a signal I'm hearing more and more often, which is: what has been disrupted? Trust has been disrupted. So you have to lean into the trust layer. It is hard to get your product listed on Apple's App Store. It's expensive, they take their cut, all of these things. But I never worry about downloading anything from there, because I know it's been vetted. That vetting and verification layer is built on trust. So with skills, that's a really interesting point. You're attacking the application layer with third-party validation: you can trust this, it will not inject malicious code. However, you are also taking on a great deal of responsibility. How do you balance that trade-off? What do you want to take responsibility for, ultimately, in this space?
That's a great question. I think we at Gen are very well positioned to do this, in part due to our history. One of our largest business lines is LifeLock. LifeLock is an annual subscription that lets you know if there are any fraud alerts on your Social Security number or any of your important personal and family information. When that information is found online, we help you remove it. As part of this, we have plans that naturally involve insurance. So we have great relationships with a bunch of insurance providers that specialize in the digital space, and we're actively in talks about how you insure something when an agent, whether it's on the attack side or on the user side as the defender, takes an action that causes some loss of value or peace of mind for the consumer. So this is something we're actively thinking through and working on.
What we were actually getting to is that, with the speed it's changing, you can't plan five years ahead right now. So how are you doing your planning for which opportunities and which threats you want to take on?
I think the way we've developed our team is a culture of speed and extremely fast reflexes. Think about when Google was really big on its X moonshot factory and had multiple bets happening at once. We run the AI Foundry in a very similar fashion. We hire a number of ex-founders who are ready to apply full founder flexibility to their day-to-day operations, and we deploy them across a wide number of bets. The idea is that a few of these bets will work out. It's impossible to predict the future, but just like in venture capital, you need a number of bets, because some of them won't work and some will, and the ones that do, because you capitalized on them ahead of time, kind of return the fund, if you will. In our case, what we really care about is making sure we deliver value to our end consumers.
How big is the unit that you're running?
It's about 20 folks.
So you're operating almost a skunkworks, I imagine, within the larger organization. And there's no 40-year-old who's had 50 years in gen AI. It hasn't been a thing that long, so you have as much experience as anybody in this space. Your background between Bain and Palantir gives you insight into both the business and managerial side of this sphere and some of the AI opportunities out there. What has building this unit been like?
Our lead, the Chief AI Officer, and I spend a lot of time thinking about how best to build this. We think there are a few bets, a few directions we want to go. One, to begin with, are startup-style bets, where we pursue specific product lines and specific opportunities for our consumers, and they are intentionally widespread. We have AI products out for families right now. We have something built for travel, and something for students. Then there are things outside of the product layer.
Are those products you're building in-house, or M&A products you're acquiring from outside?
These are all in-house. So there are a number of bets there. But even outside of the product layer, we're thinking about how we disrupt the way consumers interact with AI products and applications today. Are there things consumers are really missing? For example, context, or being able to prompt AI correctly. These are all inputs consumers have to give today, and not every consumer is on a level playing field, but they result in outputs that may or may not be desirable. So these are all things we think through. Back to your question, there are these net-new build-outs, and there are also efforts within Gen to modernize and inject AI into some of our huge business lines. These business lines are doing billions of dollars in revenue, so any step change in operations matters, whether it's better customer satisfaction through AI-powered customer service agents, or, when we think about LifeLock and insurance, how we inject AI there. We think there are opportunities within the core business for sure.
But those are operational improvements, I imagine. It's not about injecting AI for the sake of AI into LifeLock. It's more about how you improve operational efficiency or the consumer journey or UX, things of that nature?
Totally. We also think there are new TAMs we can go into. I think the example would be insuring AI actions.
What do you mean by that? That's interesting.
This is the piece where we have a strong structural advantage, having relationships with all these insurers in the digital realm. No one today can really insure the outcome or the action taken by any specific agent. So if you ask an agent to do some basic task and it leaks your Social Security number, and as a result some scammers use that to commit financial fraud against your account, there's no accountability. No insurer is going to say, okay, do we hold the agent accountable at that point? But you gave the agent the instruction. So at what point in the value chain does the insuring start or stop? I think this is a fascinating space.
That's very interesting. I think about this as the Tron problem. In the movie Tron, and you're too young for Tron, but in the original, the characters in there are the digital representatives of somebody outside, but they are that person. Identity and responsibility for agents. There's an argument to be made that every agent needs to be tagged to a real person, because ultimately responsibility has to flow from somewhere. And to your point, no one's built that yet. If we simply have billions of agents running around that don't actually answer to anybody, well, you can't punish an agent. They don't care. The buck has to stop with someone. So how do you build the buck layer? And if you're talking about insuring something, well, once money has entered the chat, we will find out who is to blame.
Yeah. And to the point of money, Gen completed an acquisition of MoneyLion, a challenger bank, for just north of a billion around two years ago. So we have a very clear sense of how consumers think about financial wellness, spending money, and how funds flow in and out. We think having all of these building blocks together before thinking through how AI will affect consumers sets us up really well. The analogy I want to give is what I witnessed at Palantir with its ontology philosophy and how that benefited it in the AI wave. If you think about Palantir and where its valuation has been historically, it was very flat after its direct public offering. Then, right after ChatGPT came onto the scene, Palantir's valuation started exploding, and it was directly correlated with revenue exploding, because enterprises really wanted to use Palantir. Pre-ChatGPT, we viewed ourselves at Palantir as one of many solutions in the market. We often indexed ourselves against the likes of Databricks or Snowflake as a data warehouse for the enterprise. Post-ChatGPT, Palantir views itself, both from a sales perspective and internally, as not having true competitors. The reason is that Palantir invested in the ontology system extremely early, and it just happened to be a really, really great system for AI to coherently take actions in a structured, harnessed way, all the way through to business outcomes. You can say it was luck, or you can say it was ingenious planning by the inventors of the ontology. But similarly, I think Gen has a really strong right to play. We completed the MoneyLion acquisition and brought financial wellness into our portfolio. We have LifeLock, which does a ton of insurance and a ton of monitoring. And obviously the big one in the house is Norton, with its decades of experience as the frontier layer of security for consumers. All of that combined, plus our reach of 500 million consumers across 150-plus countries, we think is a really strong formula for success. In terms of the path to get there, it is uncertain. Our belief is that the model players, and even the chip makers, who are getting into developing their own models, are the dictators of the direction AI will go. We're unable to predict or dictate that. The best we can do is be very educated, take bold bets, be thorough in our thinking, and hope that some of them work out and really benefit our end consumers.
Well, the products your company has amassed all speak to the fact that there are things outside of your control. So control what you can. The frontier models are going to do what they must. So you make bets where you can. Some will pay off, some won't. Nvidia was not positioned to be the most valuable company in the world before the data center build-out in the AI boom. They made some bets, and they turned out to be the right bets at the time. For you, the bet is that people need to understand what is happening to them from forces outside their control in order to exercise some semblance of control. People always feel better, not necessarily with more information, but with the right information, that they will be protected. You're trying to provide peace of mind, which is in limited supply right now. I'd like your opinion on this from the AI perspective specifically. I've felt increasingly that one of the problems is that we don't know the medium of AI. The medium dictates function. There's the old Marshall McLuhan line about the medium being the message. And the medium of AI is still very much uncertain: text versus spoken word, the product it's being used in. ChatGPT is a different medium than the Claude I'm using in my car. So what mediums do you personally think are the most opportune for you to go after right now?
Digital is in our name, right? Our full name is Gen Digital Inc. So naturally we gravitate toward digital interfaces where we're already able to reach our consumers, or new digital interfaces where we think consumers will go. That being said, you brought up a really interesting point about hardware, and how that's probably going to be a huge part of consumers' lives. If we think about the way digital experiences grew among consumers, it wasn't until the iPhone was widespread that all the amazing consumer applications and experiences were really able to thrive. Think about Uber coming out, and eventually Airbnb. It's difficult to imagine those experiences not living on an iPhone or a hardware-backed interface.
They don't exist without the medium being invented for them.
Correct. So we definitely anticipate a wave where some sort of hardware, or maybe more than one component, is deeply intertwined in many consumers' lives. And with that comes a whole wave of application-layer opportunities, so we're readying ourselves for that as well.
You still end up being at the mercy of the hardware. Is it the new cameras being put in AirPods? Is it Meta glasses? The hardware medium is the platform, and we're waiting for that platform. Gen Digital, you're making big bets. Are any of those in the hardware space?
Not on the roadmap today, but never say never, right? Again, at the AI Foundry we move very quickly, and we're constantly in talks with the people who set the pace, the big labs and the players there. We see how hot hardware has been, especially this year. Who knows what will happen? We keep an open mind, but it's not on the roadmap right now.
How big are you guys, the total company?
We're around 3,000 to 4,000 folks.
What kind of headcount growth or reduction have you seen in this AI era?
You're talking about amongst peers, larger companies?
Large companies in general. It feels like the layoffs have certainly slowed down from the peak. But you operate in a different space. You have a fair number of software engineers who I imagine were affected by this. Do you anticipate the company continuing to grow? Are you in a wait-and-see, we're-not-losing-anybody phase? How's it feeling these days?
Probably more wait and see. And we think that's pretty indicative of the macro environment as well. As you said, there was a phase earlier this year, in January and February, with huge mass layoffs. Block made quite the stir with an almost 50% reduction in workforce, claiming AI could fill in with all those efficiencies. And the market rewarded them for that from a public valuation perspective. Then a bunch of companies followed suit to hop on the trend. But we think it's overdone. All companies have done some RIFs. One of the philosophies we have within the AI Foundry is that all employees should be AI native in this day and age, because AI is so accessible. If you aren't using AI, someone else is. There are finite resources, and you're competing against other people for the consumer market. So if you're not using AI all the time, you're naturally behind. With that added efficiency, reductions in workforce are natural, and we've done our fair share this year. That being said, I do think we're in more of a lull. We talked earlier, before entering the room, about tokens, and how at some point earlier this year all the companies were saying, let's go tokenmaxxing. Then you saw the CFO step into the room after a quarter or two of that and say, hey, this is not really sustainable. So, in conclusion, RIFs and reductions will happen with AI, but we're definitely in a slower period, more of a return to normalcy.
I think the bigger question everyone's trying to grapple with, and you're in the Valley, the center of both the creation of all the new tech and the hype cycle: Bernie Sanders came out today and basically wants a moratorium on new gen AI development. There is a great deal of pushback and a great deal of anxiety in the country right now. How much of that is due to AI specifically versus broader cultural trends, we'll define five years from now when we look back on this era. But you're working at a foundry that is specifically devoted to moving AI forward. Obviously you think AI is a net benefit. Can you empathize with all of those folks out there saying, you've created an infinite layoff machine, why should I be down for this right now?
Totally. This is a big thing all AI leaders are thinking and talking about these days. Especially for us, because our end customers are consumers. If consumers have a negative perception of AI, no matter how delightful these experiences are, they won't want to use them. So this is something we're grappling with. It's probably a bit outside our direct control, more about how the kingmakers are dictating these experiences and how they're conveying the news. But to your point about understanding how consumers are feeling, we do a ton of user interviews with our existing customers and a ton of user studies to understand how the average consumer, outside of the Valley, in the US and abroad, really thinks about AI. And to your point, it's not always positive. So we think the winners in consumer AI are the ones that understand consumer sentiment around AI really well, capture the points consumers really care about, and then deliver value through those channels. And I really believe that with some sort of hardware intervention, that whole cycle might change, like the iPhone analogy. But time will tell.
We anthropomorphize things. We can fall in love with an iPhone or with a company. The kingmakers have done a terrible job marketing themselves to the rest of America. Altman and Dario have not done themselves any favors with a lot of their PR appearances. This is the first step of a two-part question: do you think your company and your product make the world a better or worse place?
Definitely better, especially as it concerns consumers. Everything we build is around digital freedom, and that extends into the AI realm as well.
And do you think AI and the technology rolling out now make the world a better or worse place?
I would argue better, but with caveats. There's never an only-positive side to any story. I think AI will make humans more productive and will make learning so much better. If you rewind through history to how humans came to dominate the earth as the forefront species, a large part of it was our ability to communicate and pass down learnings through generations. AI has completely changed the way humans can learn, in a way that has never happened in the hundreds of thousands of years we've been around. It's always been through telling, through experiences, and then at some point preserving that in writing, which we go and consume. But a human eye and brain can only process words and concepts at a certain speed, biologically and physically. With AI, the information we've gathered throughout all of time is now encapsulated a prompt away. And with some of the hardware advances people are making, you mentioned Meta glasses, the information is even more readily accessible at any given time. I'm sure there are going to be advances with what Elon is doing with Neuralink, implanting information directly into the brain. Overall, I think it's positive for humanity, for the average human, again with caveats. And Gen would love to be one of the players that delivers the right products to defend against those caveats. Whether it's, hey, I'm worried AI has really expanded my attack surface, how can Gen help? Or, hey, I'm worried I'm going to lose my job, how do I think through this? Gen has spent a lot of time on this with our history outside of the AI landscape. The reason MoneyLion became so big is that its primary product is early paychecks, and the ability to help consumers with that. So again, the world is moving quickly, and we have to move faster to stay afloat.
Look, there's never been a new technology introduced that didn't cause disruption. Things end well, but there is always disruption along the way. I think the issue we're all facing is: do we have the cognitive capacity to process change and integrate it at the speed it feels like it's happening? And is change happening as quickly as it feels to those of us on the inside versus those on the outside? If you're still working a job that operates the same way it did 10 years ago, you're simply hearing about this and being told about it, but you don't know.
Yeah, 100%. On the point that AI can't touch everything: as much as those of us who live in the Valley want to believe it can, it doesn't touch everything. It doesn't touch every single career type or every interaction humans have with each other. So I think it's good that, as you mentioned, governments are becoming more active in thinking through what to do about AI. Not just from the regulation standpoint, which has to do with geopolitics, and that's a whole other topic. But the question of how you protect the average person, how you give the average consumer or citizen some downside protection as it relates to AI. Those are complex and nuanced questions for policymakers, but it's a very needed discussion.
Well, not just the policymakers, but also the people at labs like yours, the people trying to push the boundaries. Are you operating from a position of stewardship or a position of value extraction? That's ultimately the question. How do you feel you're approaching it?
100% stewardship. And I wouldn't characterize us as a lab. We're a foundry. We want to build AI products, but we're not in pursuit of AGI, which is what all the labs are after: let me get to the best AI as soon as possible.
Oh, that's funny. I use "labs" fairly broadly, but it's true. A lab is someone specifically working toward AGI.
Correct. Whereas we're the opposite. We weren't founded with the premise of doing AI at all. We were founded decades ago on the mission of protecting consumers and giving them freedom in their digital lives. And we want to make sure we're there as consumers enter the AI era.
Is there anything you're seeing that you don't feel is getting enough attention in the broader market right now?
That's a good one. Probably data, really. Even as we're doing this recording, a ton of data is coming in and out. Data is obviously the oil of AI, and what's required to make these models better. Traditionally, folks have thought about data as online data. We see a lot of lawsuits where the labs have supposedly scraped the online corpus and infringed on copyright to train their models. But if we both believe robotics will be a big thing for consumers, how does all this real-world data get captured? This is already very prevalent in the Valley and has been very hot for the last few months, but I wonder how the rest of the world will think about it, because that's when every single thing you do, every action, could be commoditized or commercialized without you realizing it.
Every conversation you have with anybody is filled with far more data than just the words you're using: the tone, the glint in your eye. That's why when they're putting cameras on AirPods, that's as much about collecting more data to train these models. Words are such a small part of the human experience. And I was interviewing someone working on neurochips for the brain. Once you can access my thought process, I'm talking to you right now but thinking about a particular emotion at the same time, and once that can be articulated through a model, that provides color, the context of what makes us human. AI isn't anywhere close to being able to absorb that yet, but we're working on enabling it.
Correct.
So, I don't know if that's a good or a bad thing, but we're doing it.
We're doing it, and I don't think people, consumers especially, especially outside the Valley, are paying attention to it or are really aware of it.
Are you more hopeful or more anxious about the next five years?
I'm an optimistic person. I like to carry myself that way in life and through everything, so I'll say optimistic, positive.
Be part of the
conversation.
Whether you're a CTO who wants to be featured, a company looking to sponsor, or an engineering leader wanting a seat in the room — there's a place for you here.