How should we approach chatbots' mental health and privacy concerns?

Summary

TechTank Podcast Summary
Episode: How should we approach chatbots' mental health and privacy concerns?
Host: Josie Stewart (guest host, Brookings Institution)
Guests: Shay Gardner (Policy Director, LGBT Tech) and Sydney Silvestro (Senior Policy Analyst, New America, OTI)
Date: April 20, 2026

---

Overview of the Episode

This episode of the TechTank podcast delves into the complex intersections between chatbots (driven by generative AI), mental health support—particularly among youth—and the privacy and safety risks these technologies pose. Host Josie Stewart, joined by Shay Gardner and Sydney Silvestro, explores both the visible and less visible harms, the ongoing policy debates (including bans), and future pathways to responsibly shaping such tools for vulnerable communities.

---

Key Discussion Points & Insights

1. Visible and Invisible Harms of Chatbot Use in Youth Mental Health

• High-Profile Cases and Concerns

  • Shay Gardner addresses publicized tragedies where adolescents sought mental health support from chatbots and, in some cases, were affirmed in self-harm ideation.
  • Gardner cautions against cutting off vulnerable youth from the only support they may feel safe seeking, especially LGBTQ youth, emphasizing the tightrope between harm and access.

    “Those cases are absolutely moments of failure. Right. But two, we need to be very careful not to answer one failure with another...There is a reason they are turning to these systems for support in the first place.” (Shay Gardner, 02:25)

• Less Obvious Harms and Human Elements

  • Sydney Silvestro expands the lens from technological to human impacts, focusing on “mental autonomy” and how young users’ identities and beliefs may be subtly shaped by chatbot interactions.
  • The design of chatbots encourages validation (sycophancy), mirroring users’ language and emotional states, lacking the friction or fact-checking of human support.

    “AI is the mirror that smiles back, right? Like, it’s meant to mirror your language...It is meant to make it seem like it’s something that is engaging and is reflective of you, but it’s also meant to kind of, like, flatter you.” (Sydney Silvestro, 04:14)

---

2. Intersection of Bias, Discrimination, and Mental Health Harms

• Two-Stage Problem of Bias

  • Gardner outlines:
    1. How societal biases can make chatbots “feel” safer for marginalized youth than real people.
    2. The risk of chatbots delivering affirming or harmful responses, with no way to take back a damaging answer in a vulnerable moment.
  • Personal anecdote underlining the point: Gardner reflects on seeking answers about sexual identity online as a teen, noting today’s youth might ask a chatbot as their first resort.

    “Whatever that immediate generated response is, there is no taking that back...a bias response is harm. It is going to hurt.” (Shay Gardner, 08:38)

• Perceived Objectivity and Youth Vulnerability

  • Silvestro discusses the dangerous assumption that chatbots are “neutral” and authoritative, making biased responses potentially more damaging, especially for youth who lack media literacy.

    “The assumption that these answers are objective, even when they obviously reflect bias of the data they were trained on. You need to have a lot of sophisticated digital literacy to understand that.” (Sydney Silvestro, 11:02)

---

3. Social and Developmental Contexts for Youth

• Changing Nature of Relationships

  • Gardner underscores the shift from parasocial (influencer/audience) relationships on social media to highly personalized, frictionless one-to-one relationships with chatbots.
  • This increases the blurring of lines between real and artificial intimacy and support.

    “What makes chatbots different is that they feel one to one...it makes the relationship feel more private. We know it makes it feel more personal.” (Shay Gardner, 15:48)

• What’s Being Replaced?

  • Both guests agree that research should focus not only on chatbots but on systemic issues driving young people to seek support online (e.g., underfunded school mental health services, weakened “third spaces” post-pandemic).

---

4. Privacy Concerns and Digital Literacy

• Chatbots and Intimate Data Collection

  • Gardner points out the critical difference between a chatbot and a human therapist: there are no confidentiality rules for chatbots. This creates an environment for “quiet collection of incredibly intimate data” without users’ informed consent.

    “No matter how much a chatbot may feel like a therapist...it is not bound at all by the privacy or confidentiality rules that would apply to an actual therapist.” (Shay Gardner, 22:24)

  • US privacy law remains “far behind,” lacking a comprehensive federal privacy framework for youth or adults.

• Privacy as Agency and Precondition for Safety

  • Silvestro reframes privacy not just as secrecy, but as agency and control over one’s data and identity.

    “Privacy is like a precondition for safety. I really believe in that...it’s about the ability to learn, it’s about the ability to fail...without all of these different inputs telling you who you are and aren’t.” (Sydney Silvestro, 24:24)

---

5. Policy and Regulatory Approaches

• Fragmented Federal and Legislative Landscape

  • Current US policy is complicated, fragmented, and often conflates or separates safety and privacy in incomplete ways.

    “We are certainly lacking some sort of comprehensive framework...it is still responding in a very fragmented way.” (Shay Gardner, 28:46)

• Debate Over Bans and Age Restrictions

  • Both guests strongly argue against blanket bans of chatbots or social media for children and teens, warning of negative consequences (cutting youth off from support, disproportionate impact on marginalized users, pushing users to less regulated spaces).

    “It is a categorical block between an individual and lawful online participation and speech. Two, it is absolutely going to have a disproportionate impact on marginalized or isolated youth who are in non-traditional or harmful situations.” (Shay Gardner, 34:44) “A ban is always, it’s never going to, it’s one, it’s not going to work and two, it’s going to create so many terrible consequences that it’s really not worth it in my view.” (Sydney Silvestro, 36:45)

• Digital Literacy as a Solution

  • They advocate for building “friction” into systems, increasing user agency and critical digital literacy, rather than exclusion.

---

6. Product Design, Future Hopes, and Positive Use Cases

• Responsible Technology Design

  • Stewart and the guests discuss the importance of industry moving toward products that are evidence-based, designed for quality of engagement over “time spent,” and attuned to the real needs and circumstances of young users.

• Supporting Safe, Honest, Limited Use

  • The hope is not for chatbots to become substitutes for human care, but to act as tools that help youth access real support, information, and reduce feelings of isolation or stigma.

    “My biggest hope is not ever that chatbots become stand ins for human care. It really is that they become safer, more limited and ultimately more honest tools...that can really help young people find the information and the community that they feel like they are missing...” (Shay Gardner, 41:16)

• Nuanced, Evidence-Based Policy

  • Silvestro calls for nuanced, context-aware policymaking and research, and highlights promising developments like evidence-based mental health chatbots (e.g., Dartmouth’s "Therabot").

---

Notable Quotes & Memorable Moments

• "Welcome to the tightrope." (Shay Gardner, 03:14) —On the balance between harms and needs in offering online support.
• "AI is the mirror that smiles back." (Sydney Silvestro, 04:44) —On chatbots’ tendency toward validation and flattery.
• "Data doesn’t just happen, is what I always like to say. It is created." (Sydney Silvestro, 14:05)
• "If we only study the chatbot and we don’t study the conditions that are pushing young people towards it, we are missing a huge part of the story." (Shay Gardner, 15:48)
• "Privacy...is like a precondition for safety." (Sydney Silvestro, 24:24)
• "It feels a little bit footloose to me. A little bit of ban the dancing altogether." (Shay Gardner, 32:44) —On the reductiveness of bans.
• "Regulation has the ability to consider nuance and I hope that it does in that sphere." (Shay Gardner, 41:16)

---

Timestamps for Key Segments

• 02:25 — Overview of youth chatbot use and initial reactions to tragic consequences (Gardner)
• 04:14 — Less visible harms: mental autonomy and chatbot sycophancy (Silvestro)
• 08:38 — Bias and discrimination in chatbots, impact on marginalized kids (Gardner & Silvestro)
• 15:48 — Differences between social media and one-to-one chatbot relationships; research needs (Gardner)
• 22:14 — Privacy as a core concern, especially for youth (Gardner & Silvestro)
• 28:46 — Current policy/regulatory landscape overview (Gardner)
• 30:24 — Problems with bans and age restrictions for youth (Silvestro & Gardner)
• 39:17 — Hopes for the future: positive uses, responsible design, evidence-based policy (Silvestro & Gardner)

---

Conclusion

This episode underscores the nuanced, multi-layered risks and opportunities that chatbots present for youth mental health and privacy. The guests advocate for research and policy that center real-world needs of vulnerable communities, promote digital literacy, prioritize meaningful privacy, and resist reactionary bans that risk greater harm than good. Instead, targeted regulation, responsible tech design, and social investments form the path forward for safe, empowering digital futures.

Transcript

[TechTank Host]

You're listening to TechTank, a biweekly podcast from the Brookings Institution exploring the most consequential technology issues of our time. From racial bias and algorithms to the future of work, Tektank takes big ideas and makes them accessible.

[Josie Stewart]

Welcome to the TechTank podcast. I am today's guest host, Josie Stewart. I'm a senior research and communications assistant for the center for Technology Innovation at the Brookings Institution. Many of the big AI companies have promoted their large language models across a variety of use cases, touting how generative AI can be used beyond work for more personal use cases such as health questions or companionship. Survey data shows that users are experimenting with chatbots across these domains. Though the uses differ by age, children are regular users of the technology, using chatbots for everything from homework health to mental health support. In fact, survey research indicates that about 1 in 8 U.S. adolescents and young adults have used generative AI for mental health advice. This use has raised concerns among lawmakers, especially following the deaths of multiple teens by suicide. After chatting with the models, researchers are now trying to quantify concerns about safety and effectiveness. And beyond actual mental health support, there are additional questions on how these chatbots respond in suicide, serious mental health crises, and can obscure privacy risks for users. Today I am pleased to be joined by two guests, Shay Gardner and Sydney Silvestro. Shay is a policy Director at LGBT Tech, where she leads the organization's policy strategy and research on LGBTQ digital rights and online safety. Sydney is a senior Policy Analyst at New America's Open Technology Institute. Her work focuses on privacy and data policy that keeps emerging technology safe and beneficial for vulnerable communities. Thank you both for joining me today.

[Shay Gardner]

Thank you so much.

[Sydney Silvestro]

Yeah, thanks for having us. Really excited. Great.

[Josie Stewart]

So I wanted to start out with more of the, you know, the most visible harms we've been seeing dominate the headlines from these models. Shay, can you walk us through some of the tragic consequences we've seen following teens use of chatbots and what were your initial reactions to these news stories given your work at the intersection of technology and vulnerable communities?

[Shay Gardner]

Yeah, I would say the most visible harms, really the most headline making cases, right are the ones where a chatbot is being specifically experienced as this one to one complete substitute for a real world person. Point of support, whether that's a therapist, a friend, a romantic partner, and I think we've all seen in the most tragic cases, you know, these systems go as far as affirming or validating self harm ideation in the interest of being systems that are as affirming and validating as possible. So, quite honestly, I. My initial reaction to that is twofold. One, those cases are absolutely moments of failure. Right. But two, we need to be very careful not to answer one failure with another. If we are watching young people become too reliant on these systems for support, we do have to recognize that there is a reason they are turning to these systems for support in the first place. Everything I talk about is grounded in LGBTQ perspective from the youth, in our community's perspective, that is especially important for many young people. These tools are not replacing this rich existing network of offline support. So the answer cannot simply be to cut these vulnerable youth off from something they are using to seek connection information or support that is otherwise unavailable. So I think that's a long winded way to say my reaction now is welcome to the tightrope.

[Josie Stewart]

Yeah, that's definitely, you know, assessing the larger problem is something that we've been talking about a lot in CTI Sydney. Do you have any other thoughts on that, or what other types of harms are we seeing that might be less visible but still equally as harmful?

[Sydney Silvestro]

Yeah, so I think that everything that Shay said was spot on. And often when people want to talk about, like, AI or chatbots, they want to talk about the technology or they want to kind of like focus in on the technology. And I think the thing I always try to center both in my work, but also in these conversations is like, these are really the conversations that we're having about humans and how humans interact with technology. So it's a story about people. There's absolutely tech failures here. There's also ways in which the tech is beneficial, but it's really about, like, especially in this context, how young people are navigating their emotional world, how they're making sense of themselves, how they're making sense of others, what tools are they using, as Shay said, to really fill some of those voids that they're experiencing or to try and figure out where they fit in the world, whether or not chatbots should be the thing that they use for that. That's exactly. I think that's part of the tightrope that we're trying to navigate, as Shay said. But there's still this, the nature of the interaction. There's still some human element to it. And so I think that's where we really need to hone in on what that looks like. Some of the kind of components that I think have not been quite as visible, but are still really important. We can talk about privacy all day. But outside of the privacy realm, which I think there's kind of two components that I really have been thinking about a lot. So one is the mental autonomy. So Nita Farahani has this call for cognitive liberty, but really, I think when we think about young youth users who are still developing their identity formation, who are still kind of learning how to think critically, how these kind of, like, conversations, this relational approach to AI, can subtly shape their beliefs, their emotional interpretations, their perception of reality, and the kind of, like, what are the legal protections that are in place when we're thinking about that, where that. What I really mean is a lot of the kind of data extraction, some of the lack of privacy around that. And the other bit, which I think has come up a lot more, is this kind of kind of, like, sycophancy. So the way I kind of, like, put this is that AI is the mirror that smiles back, right? Like, it's meant to mirror your language. It is meant to make it seem like it's something that is engaging and is reflective of you, but it's also meant to kind of, like, flatter you. It's by design. It's very much focused on, like, validating because that is part of how the models are built. And this is, like, to be, like, clear, like, we're acting like this is outside of the realm of what we've experienced before, but it's actually, this is something humans have been doing for a long time. We flatter each other. We seek validation from each other. There was an essay that Plutarch wrote in 100 AD that was titled how to Know a Flatterer from a Friend. So this is something we've been dealing with well outside of the realm of technology. But I think there's something to be clear, I think there's something in the way people are interacting with ar, the way that AI presents itself, that is different. And that really has to do with not necessarily causing any of these delusions or anything like that, but really more kind of reinforcing it. So I think when we're thinking about paranoia, mania, those type of things, when you're interacting with a human, there are these kind of friction points where that human might say, whoa, let's check the facts, or let's stop here and understand what's going on, or here are some source that you might have access to. When we don't build those into a chatbot, that doesn't happen. And so it's not necessarily necessarily that the chatbot is the reason why people are delusional. It's just that you don't have that same kind of, like, human friction. So those are like, some of the areas that I think are concerning. But on the other side too, and we can talk about this more too, is like, there are also a lot of, like, really beneficial uses of AI and so don't necessarily want to basically say that, like, we can't solve this problem. And so our best approach to solving is just to, like, ban kids.

[Josie Stewart]

Yeah, I like how you talked about kind of the problems, you know, that are. We're seeing with the platform design or the way AI is built. I think one of the bigger conversations that have, you know, we've been having for a long time is bias, discrimination in models. And I'm kind of curious, Shay, if you can talk about those, are often treated as, you know, separate issues. But where do we see, like, bias and discrimination intersecting with mental health harms? And especially given your work, you know, what impact can this have on kids with marginalized identities? Given everything Sydney was talking about with the ways these specifically can affect children?

[Shay Gardner]

Bias and discrimination is both one of my favorite things to talk about and one of the most frustrating things that I continue to continuously have to talk about.

[Sydney Silvestro]

Right.

[Shay Gardner]

But I do think, I think there are two stages to this when it comes to bias and discrimination in this context. So to ground my first, right, none of these experiences are happening in a digital vacuum. These are systems that are built by people. They are trained on human language, and at the end of the day, they are being deployed into a world that contains a whole lot of very bad and very, very real bias and discrimination. Technology is not creating those dynamics from scratch. And I think that is always very important to remember. What it often can do is replicate them, it can scale them, it can amplify them. And it is on the developers and the deployers in that case to sort of push back on that element. But I said there were two stages. So this is my question for the first. Have we built a world where bias and discrimination is going to make chatbot feel safer or more accessible than a person? The second stage of this, I actually, I like to tie my own personal experience into it. I was 15 the first time I used a web browser to search the term is it okay to be gay? It was a very real world of bias and discrimination that made me feel like I was going to be safer. Seeking out what I thought at the time was. I know probably wasn't. But what I thought at the time was going to be an anonymous or at least A lower risk source of support. If I'm being honest, if I were 15 today, I would probably go ask a chatbot. And whatever that immediate generated response is, there is no taking that back. Whether it is affirming, whether it is the opposite, there is no taking back what that does to that young person. And I promise a young person in that moment is not looking for and definitely not in need of this, this neutral breakdown of competing ideological positions on queerness. So it's. I said the first part of the conversation is, have we built a world where bias and discrimination make the chatbot feel saf than the real world? And I think the second question we need to be asking ourselves about chatbots is have we built the chatbot to answer, well, once the young person gets there, if we are talking about protecting young people from harm for marginalized kids, a bias response is harm. It is going to hurt.

[Sydney Silvestro]

I want to pick up on what Shea said there about the neutral breakdown of these potentially, is it okay? Is it not okay? I think that the bias and discrimination point, this idea that it's a neutral technological approach to it, right? Like when you go ask your older cousin and they say something and you have all this contextual information about where they might be coming from, and even if you're like 10, 11, 12 years old, you might still have some sense of they are an external entity from you and it might hurt, right? Like whatever they say, whether it's like you're talking to them about your gender identity or body dysmorphia or even just like what you want to be when you grow up, and they're like, you could never be a doctor. Which, by the way, is something that, what said to me, the, the, the kind of like that response from another human hits differently than it does when it's coming from something that you're kind of taught to interact with in a way that makes it seem authoritative and technical and quote, unquote, like, neutral, because it has this faux perception that giving, you know, two sides is the most important thing. And I think that that like the assumption that these answers are objective, even when they obviously reflect bias of the data they were trained on. You need to have a lot of sophisticated digital literacy to understand that. And you also need to have a lot of emotional resiliency to separate out from that. If you are trying to get diet advice from. I definitely, I don't know, I had Seventeen magazine. There were definitely. There was a lot of bad diet advice. There's still a lot of bad diet advice on the Internet, and you're going in and you're asking these things and you're saying, well, sometimes I feel dizzy and I. I want to make sure that I am losing weight. And the chatbot is picking up on that. And it will just pull from all of this really terrible, circa 2012 diet advice from Cosmo. I think Cosmo was still a thing in 2012, whatever it was, where it's like, just drink more water. And if you don't know all of the elements of how that data has been formed in, and it's also not coming from a person, it's not on a forum where you have the sense that you're engaging with another human, that it can hit differently. And so I think that's something too, that with like, the bias component, it's not just that the information that we're receiving is bias, it's also that our sense of it is that it shouldn't be.

[Shay Gardner]

Absolutely, yeah. I think this also speaks to. When you're looking at the way regulatory, legislative approaches to AI chatbots and to these, you know, technologies like these,

[TechTank Host]

I

[Shay Gardner]

think this is a perfect example of why you cannot look at that with sort of broad moral authority, strokes of legislation or regulation, where some individual at some level of government has the authority to decide what is or isn't either neutral or what is or isn't right or what is or isn't harmful.

[Sydney Silvestro]

100%. And we built the Internet, like we put all that data on there. It reflects us, unfortunately.

[Shay Gardner]

Yeah. The good, the bad, and the worst. Right.

[Sydney Silvestro]

Data doesn't just happen, is what I always like to say. It is created. Yeah.

[Josie Stewart]

No, I think you both have singled in on something I find really interesting for context. I recently turned 23, so I am a child of the Internet mostly, who is Googling things as, you know, a teen. And so I'm really curious on the way that might even just affect interaction between teenagers who might not be as vulnerable with one another asking these questions, like you both were saying. So I'm curious what you might think about that, but also if there's any other questions you think we need to be asking or that researchers are still trying to answer that feed into the information we need to know to shape that regulatory environment, like Shay's saying.

[Shay Gardner]

So I do think that there is one part of this discussion that I think we could be diving a little bit more in depth to, which is how young people are coming to these chatbots. These are young people that are. Are used to building relationships and communities through screens. Right. We were Just talking about that and you know, even the development of something like a parasocial relationship is not new here. It's not new to AI chat bots. That's been around for a very long time. Young people have, have been forming attachments to creators or to influencers for a very long time. But I would love to see more research about the difference in those interactions on social media versus in AI chatbots. For traditionally the relationships between individuals on social media have, have been sort of a one to many relationship. What makes chatbots different is that they feel one to one. And Sydney mentioned friction earlier. It's a, it's a frictionless one to one at that.

[TechTank Host]

Right.

[Shay Gardner]

That changes the dynamic. It makes the relationship feel more private. We know it makes it feel more personal. Most people, the understanding is that that may increase in, in blurring some of the lines of what that relationship is. I would like to see a little bit more of that comparison search between social media versus AI chatbots. Because a lot of legislation right now is either trying to hit it both or unintentionally encompasses the other while trying to hit it one. I also think that there is a, there is a second and selfishly this is just research I would really love to see. Just like I said, bias does not exist in this vacuum of digital spaces or within these AI systems. This youth mental health crisis is also not just going to disappear if a chatbot does. So I am, I am very curious, what is that chatbot replacing? But not only that, why was that thing missing in the first place? So I, I would love to see more research lean into that question. I think understanding what is missing for young people is truly the first step to building a world where they can thrive offline and online. It is, you know, if we, if we only study the chatbot and we don't study the conditions that are pushing young people towards it, we are missing a huge part of the story.

[Sydney Silvestro]

I think that's like spot on again, which I'll probably keep saying. But the me too, these things, again, they don't just happen in a vacuum. Right. Like it would be part of the reason why I think there's such a focus on trying to solve the kids online well being thing is that it feels like it is both hard and difficult and awful. And it also feels like an easier thing to solve than some of the real world harms that kids are facing in the real world. Right. Like that are systemic and that have all the different components and institutions that are attached to them. People to separate out the like real world from the online world. And I just, I think they're part and parcel.

[Shay Gardner]

Right.

[Sydney Silvestro]

Like when we talk about cyberbullying, we still also have to talk about like in person bullying. Right. And there was a lot of work that was done to kind of shift away from this framing of cyberbullying as a thing that was separate. I think what she said about like, what is driving young people to these platforms. Some, some, some of the examples that, you know, like we can point to, it's like there was the Trump administration cut $1 billion in school based mental health support for K12 students in 2025. Right. That that was actually put in place after the school shielding in Uvalde and they cut that. And it's like we already know that there's not enough third spaces. We already know that all of these different ways, these different places where kids have been able to form relationships and get to know one another, we know that shifted during the pandemic. And we're not necessarily trying to build that, but instead we're just kind of saying like, well, the best we can do is ban them from the one space where they're potentially finding some type of interaction. I think also what she said about like this one on one thing, it's, people want to continuously compare, I think chatbots to social media and they want to say kind of like, well, we missed the boat on social media and so we're, you know, we can't miss it on chatbots. I think that's true. And I don't think, I think that what is concerning to me about chatbots is that actually the type of interaction that you're having. What she said about like this one on one component, it's very different, right? Like social media. Yeah. You have all these people in your network and you have influencers and you need to understand how they're trying to sell you some things. And you need to understand that, you know, there's a certain element of what's being presented is not actual reality for everyone and might make you feel lonely. And all of these other things that we know have impacts on people generally, not even just young people, but people, but with a chatbot, this kind of relationship that you build. And it's really hard for humans not to anthropomorphize things too. This is the other thing. If we could have a dispassionate relationship with some of these chatbots, which some people can, this would look different. But it's people very quickly become like kind of attach some human elements to the things that they interact with. But also this is being done intentionally by these companies to make them feel like more palatable or more friendly. There was an interesting report that came out that was looking at how AI toys for like young kids might actually shift some of their early learning development. Because it's all about building that one on one relationship. And when you're at that age, like 2 to 3, a lot of it is about understanding how you fit in within a group. And it's really like an outward facing, like developmental stage. And so instead if you have these toys that interact with you, it becomes much more interpersonal with that, relational with that toy and what that does to kind of like where we're going with these things. So what really want to see is some element of what is different about this technology and then how is it impacting people? I would love to see that before the tech is released. I feel like with like character AI, they basically were like, we're just going to put this out there and we'll figure it out. And then it had some, you know, impacts that were not great. And then they like rolled it all back and they said, like, we actually can't promise that the chatbot function is going to be safe enough for those under 18, so we're just going to like cut it off. The other component that I would like to see that I think is really important too is what actually works in helping people understand what this technology is doing. So the digital literacy component is huge. It's also hard, like, right, like critical thinking, all of that. Like, we don't teach it well in school. We don't always understand how a robot or a chatbot is not a human, how there's like emotional manipulation, how there's dark patterns. Like I work in all this stuff and I still sometimes I'm like, oh, let me like remember that algorithmic bias is like a thing, right. And that actually when I like Google search something and I'm like, oh, it came up at the top of my page, it's like, yeah, because it knows what I'm looking for. It's not because it's like at the top of everyone's page. So, yeah, so I think if we could stop acting like AI is magic and actually like understand how it works and what the impacts on people are, that would be really helpful.

[Josie Stewart]

Yeah. And I think something that undercuts everything we've been talking about, but especially what you were just getting out there with the interpersonal relationships, is the privacy concerns, especially when it comes to children and digital literacy and maybe not understanding what they're having a conversation with or what they're, you know, sharing with the entity, for lack of a better term. So, yeah, Shay, I'm curious if you can talk a little bit more about, you know, what are you seeing people talk about in terms of privacy considerations? And why are those so important, especially for kids?

[Shay Gardner]

I hope you did not just hear me quietly gasp and you said my favorite word, privacy.

[Sydney Silvestro]

I am, but it is

[Shay Gardner]

when you look at the way some of these chatbots are being used. I'm going to start with an example here. No matter how much a chatbot may feel like a therapist or a counselor to the person that is using it, it is not bound at all by the privacy or confidentiality rules that would apply to an actual therapist or a counselor. Without deeply understanding that, these systems can very easily become tools for what I like to call a quiet collection of incredibly intimate data about yourself. And that collection is happening without users fully understanding what data is being stored about them, how long it's being kept, whether that is ever going to be seen by humans, how that could be used against them later. None of that. So while I do, I always like to say a little bit, I think it is very important to state clearly that a person's data is theirs to hold close to their chest to share, whether that be with. With, you know, a piece of technology or with other people. But that that comes with. That has to come with the side of meaningful choice. And meaningful choice absolutely depends on understanding the privacy conditions you're facing on the other end. So privacy is a tremendous part of this conversation. I also have to say, while we're talking about privacy, we are so far behind the mark on basic privacy protections for everybody, not just youth, in the United States. We still do not have a comprehensive federal framework. We don't have an omnibus federal, federal privacy law. So every single user in the United States is operating under a system with major, major gaps in it. And it is something I never want to be lost in this conversation when we're talking about protecting this specific subset of youth users, none of us are protected right now. That young person is not protected. Not, excuse me, the privacy is not being now at the age of 15 or 16, and when they turn 18, it's not going to be protected then either.

[Sydney Silvestro]

As a privacy person, I always want to talk about privacy. I think there's. There's a point where, I don't know, I'm sure you guys have experienced this, but there's a Point when you're listening to a podcast, and whether it's like, they're talking to people who care about privacy or they're talking about people who are more on the, you know, safety component, which I'll talk about how that's a false dichotomy in a second. But they'll be like, well, privacy advocates want this. And I'm like, yeah, privacy advocates do want that. Like, there's so many things that where if we just, like, protected people's information, if we just had, like, a basic federal privacy legislation, you could solve a lot. Not necessarily solve, but you would improve a lot of these problems, and some of these problems you would actually solve. So I think this, like, the privacy safety dichotomy, like, being positioned as opposites drives me crazy. Like, loss of privacy becomes a safety risk. Right? Like, I actually came into a lot of this work because I was working with. With youth who'd experienced commercial sexual exploitation, and we were working on some foster care reform. One of the first things that I had to do was I had to go through all of the different files to understand what information we had, and they had assented to us using this information. And it was for a research study, and it was very much focused on, like, system reform. But also one of the things I wanted to do was we would do these interviews and we would do these focus groups, and we would also have more surveys. And I didn't want to continuously collect information or from them that they had already given up to people and they had already said that we could use. I wanted to focus in on the things that they felt like they had control over telling us for the purposes of this study, instead of just kind of, like, treating them as, you know, places where we could, like, get more data out of. And that, like, that point that, like, privacy as agency that gets lost so much in this conversation. Like, we always. I feel like people think about privacy, especially privacy for kids, as, like, secrecy, right? I don't know how many people grew up with this. Like, we don't have secrets in this house. Like, we tell each other everything. And it's like, well, one, that's not true, and two, it's totally normal to have secrets. It's totally normal to have some sense of control over the information that you do and don't want to share. We have, unfortunately, now live in a society where there is so much data extraction that is happening that you don't have meaningful choice and user control over that. But the privacy is, like a precondition for safety. I really believe in that and that's a huge part of the reason why I believe in privacy and why I work in this space. It's not because I necessarily want to, you know, like put my interest over others, which I feel like is sometimes how it comes across, but it's really because I care about kids, I care about vulnerable users and I want everyone to have like some sense of control which I was lacking at various points in my life over who knew what about me and how that made me feel. But also I've seen with other people and so like all the different ways in which like privacy then becomes something that is kind of intrinsic, right? Like we. It's not just about the secrecy, it's about the ability to learn, it's about the ability to fail. It's about the ability to kind of like develop your own sense of self without all of these different inputs telling you who you are and aren't. It's about like your ability to not self chill and to kind of like express yourself freely and to like. That's just like the kind of like pro social conditions around like privacy. And then there's all like the harm that's protecting you against which is like the real world safety risk, right? Like people are always concerned about this kind of like stranger danger element to it. It's like, well, yeah, but also if you're collecting information on where people are, that's also a safety risk, right? Like we've actually seen some cases where data broker have made that information accessible to people and it's led to horrible things. The data persistence, the kind of like how it builds up a profile of who you are over long term. Like how do you. I don't know, I. You know, when you have like friends who you've known for a really long time or your family, right? Like just people who've known you for a long time and you're like, that's not who I am anymore. What happens when that's the chatbot that you're doing that with, right? Where you're just like, that's not who I am anymore. Like I've moved on from that. But it doesn't let you this kind of like profile a bit doesn't let you kind of become who you want to be and kind of like have some sense of agency over yourself. So that's like, yeah, I could talk about privacy all day, but that's kind of like my. I think privacy and safety are like very, very much a part of this whole entire story.

[Josie Stewart]

Let's Turn to the legislative side of this whole conversation. Shay, can you lay out kind of what the landscape looks like at the federal level and what do you make of, you know, how, you know, safety and privacy are kind of being treated as separate entities in the bills that we're seeing introduced?

[Shay Gardner]

I would say that the landscape is moving, but probably unsurprisingly, that movement is not what I would call cohesive or very coherent. What we are watching right now in kind of this AI chatbot landscape is, is a complicated mix of FTC inquiry, of updates to existing children's privacy rules that may or may not impact their usage of AI chatbots, and a few targeted bills here and there, but ones that, in my opinion, tend to focus a little too solely on specific use cases versus tying the safety needs together with the privacy protections in the way we really need. So it's. It's a little bit of a mess. We are certainly lacking some sort of comprehensive framework, but as I mentioned before, we are also lacking a comprehensive privacy framework. And the US Government has, has not been able to land on a comprehensive, updated social media landscape for youth either. So I suppose it's not a surprise. If I have to give a short answer, it's going to be that the federal government is clearly paying attention, but it is still responding in a very fragmented way. And those fragments have, I think, landed into a policy landscape that was already fairly inadequate to begin with.

[Josie Stewart]

You both mentioned earlier also, especially Shay, you talking about social media, how there's kind of been a turn toward bans in the children's safety privacy space. Sydney, can you talk a little bit about that? And I know you hinted at your thoughts toward bans around chatbots or even social media and how those might be connected.

[Sydney Silvestro]

Yeah, absolutely. I think one of the things too, that would be helpful here is in a shape pointed to, there's some varied movement at the federal level. But one of the things that's not always clear is what are we talking about when we're talking about chatbots. Right. Like, there is a difference between character AI. There is a difference between chat GTP that is being used to help kids with homework or that is being used to, I don't know, as a translation tool. There's a difference with Siri where it doesn't have the same level of interaction. But so, yeah, I think that's like a good point to kind of bring it back to. So we've seen a lot of like, conversations around social media bands. There's some countries that are moving forward with bans for people who are under, you know, Australia is moving forward with like a band for under 16. I think the thing that I always come back to with a ban is like, what is the actual problem that you're trying to solve? Is a ban actually going to accomplish that? And also what are the unintended consequences that are related and associated with that? So I think with like a social media ban, generally what we're saying is we cannot build these to be safe enough. We are not designing for the most vulnerable. And so instead of actually trying to improve on that with things like privacy by design, safety by design, by actually, like building user controls into all of this technology, we're just going to kick off the people that are most impacted by it and we're going to call it a day. I know that's like a simplistic way of putting it, but that's sometimes the way that it comes across to me. So I don't think that, like, bans are the solution. I think that instead what we need to do is we need to put friction into the system. We need to actually give people, like, user controls, like user agency. And I also think that the other thing with bans, and we'll see this with, like, whether it's with social media ban or chatbot ban, is that it pushes people onto less regulated platforms. It pushes people onto, you know, sites that have, like, potentially worse outcomes, worse impacts. I do think that there's like, like, with some specific technology, there is probably a conversation to be had around how we actually build it and design it. So I'm thinking here about, you know, if we want kids to be able to access General Purpose LLM so that they can use it for homework or just like, for whatever, right? They have the right to kind of use it how they see fit. Do we need to make it, like, flirtatious? Right? Like, is that something that is actually, like, benefiting all of us? I don't know. I don't really know where that line is, but I think there's conversations to be had there.

[Josie Stewart]

There.

[Shay Gardner]

I, I would say when you, when you look at social media bands in particular, a lot of language of which is, Is being borrowed. When you talk about AI chatbot bands, it, it feels a little bit footloose and to me, right, a little bit of ban the dancing altogether. However, there are several obvious and immediate problems with these social media bands in a way I imagine you would probably see replicated if you tried to take that exact approach with AI Chap on. For one, it is essentially a. It is a categorical block between an individual and lawful online participation and speech. Two, it is absolutely going to have a disproportionate impact on marginalized or isolated youth who are in non traditional or harmful situations. And three, the downstream effects of this are not actually just going to be youth. That is is ostensibly a ban on any individual adult or youth who is either unable or unwilling to prove their identity or their age. So the impacts of that would be tremendous. And this is the. Looking at an issue and recognizing there are harms that need to be addressed and can be addressed on the regulatory level is one thing. Looking at that issue and saying completely banned is not something that has ever worked when it came to young people or adults and something that they found valuable to access. LGBT Tech has spearheaded a spearheaded coalition statement with a series of other organizations in direct opposition to these under 16 bans or to similar minimum age access bills. We see them pop up every once in a while on the federal level and in the states in 2026 they've been having a run at it. Hawaii, a bill has been moving through the legislature that's an under 16 ban. California has an active intent in place to consider an under 16 social media ban. And Vermont and Minnesota are also looking at it. So this is a prospective set of bills that have been moving through different states and occasionally on the federal level. And very happy to go on the record here and say this is unequivocally the wrong response to it.

[Sydney Silvestro]

Yeah, I think Shay, I think you set that up really beautifully. And one thing that okay, so this, I found this really interesting. So in Australia where they're approaching, they're doing this under 16 ban which was being spearheaded by the Esafety Commission, so regulatory body and the Human Rights Commission of Australia. So again like not a outside UN entity like within the, within their own country. They came out and they said that this actually violated kids humans rights to entertainment, to access to information, to connection. And I thought that was really interesting, right? Like there's a lot of tension here. Like a ban is not the right approach to making sure that people have access to connection, to information, to enjoyment. Like when we look at how a lot of people are using this like and there's, and as she pointed out there's like also the knock on effects of not just kids who are being like caught up in this. Right. So chat GTP I think said came out that it was going to use probabilistic age estimation to understand like what age people were and if they were estimated to be under the age of 18, then they wouldn't be able to access certain types of information, wouldn't be able to interact in certain type ways, that's still like a probability. That's not like a perfect approach to any of that. But even without all of that too, like, I think the other part with bands is if you're kicking people off of these platforms like, and you're pushing them underground and you're cutting them off from like meaningfulness, necessary support, and you're putting all the privacy considerations that go into that too, and the free expression ones also you're not preparing people to interact with these tools in the future. There's been so much chatter in the last like year about creating like an AI ready workforce. Kids are the future of work. Right? Like that is where it's going to be happening. And so I think also when you're talking about like, well, we're going to keep them off of this. There's, there needs to be a tiered approach. There needs to be a lot more digital literacy. There needs to be like, like more friction built into all these systems so that they're healthy for everyone. And as she said, like, we don't need, there are some harms that are happening here. We need to address those harms. But a ban is always, it's never going to, it's one, it's not going to work and two, it's going to create so many terrible consequences that it's really not worth it in my view. And we could do better.

[Shay Gardner]

The growing into AI is actually, that's really prescient as well, you know, not only for the AI workforce, but also the way so many of these bills treat all youth is if they're this monolith and then on the magical day they turn 18th, are imbued with an entire understanding of digital literacy and participation is just not the reality of the situation.

[Sydney Silvestro]

No 100%. And you can totally teach digital literacy. You can teach critical thinking without necessarily letting people have complete unfettered accesses. I'm just talking about within the context of learning. But I will also say I had a pretty good understanding of, of you know, research and stats and how that kind of like quantitative methods before I learned how to code and then once I did, like once I actually started doing my own research that was like really quantitative, I had a totally different understanding of it. Right. Like I understood how the data worked, I understood what this type of regression mean. I like, sorry, this is really nerdy and boring, but it did it like I Still had like, I could read a research paper before and be like, I don't know, these methods seem like a little sus. But now I can really, really break it down in my mind. And if you don't interact with these things and then yes, you just magically all of a sudden that you 18 now know how to interact with, I also think then you're also creating people that are very much like, at the mercy of AI as opposed to helping to make the tech work for humans, which is also what I would like to see.

[Josie Stewart]

Yeah, I think you both offered such good examples there that you already answered a little bit of what I was going to ask for, what measures you might find more effective. But I want to end us on a little bit of maybe a more positive note. And you both have throughout this whole thing mentioned kind of the positives and the benefits that can chatbots offer and how those are a little obscured by the bands. But what are your biggest hopes for, you know, this whole conversation moving forward? How do you hope people will be able to interact with chatbots, especially kids, and have those benefits without losing access? You know, how do you see this moving forward and what are your hopes for the technology as that proceeds?

[Sydney Silvestro]

I think my forever hope with all of these conversations is that we approach it with nuance and we kind of approach it with like curiosity to get at the heart of the issue. So really around like, what is happening to whom? When Shay mentioned people who we have a lot of assumptions around, like what type of kids we're talking about here, usually what that means means I feel like in the conversation it means like kids with parents who can actually like engage with them on this and who have, whether it's like a supportive household or non supportive household, but they still have like adults in their life. Like a lot of the kids I worked with did not have those people. And so what does that look like? And I think that to kind of approach a lot of these things as if like those people don't matter is really problematic to say the least. But so I would hope that we can have a little bit more nuance on the like, positive side too. Like, the positive side of AI is I do actually think there's some like, really promising applications. I think that at its best what AI can do is that it can help scale solutions to problems that either we don't have the resources to address or we don't want to spend the resources on addressing, which I think is a lot a second time. But like, just to take it back to like the youth mental health crisis. It is also true that there are not enough counselors, right. There is not enough, like quality therapy that people can help to help them, like make sense of all these things. It's hard being a teenager. It was long enough. No, for me, like, I'm not, I'm not quite Josie's age. It was long enough for a go for me that I was like, oh, yeah, maybe it's fine, but no, it's hard being a teenager. Like, there's so much happening there. And so I think that like, so Dartmouth, for example, is developing, I think they called it Therabot, but they're developing a AI platform that is meant specifically as a mental health chatbot. And they're actually doing randomized control trials and it's supposed to be evidence based. And so I'm excited to, I'm excited to see these like, small, very precise applications and seeing what is possible with that and how it might be able to help people. I think though, that with the kind of larger models, it's always going to be a little bit tricky to balance between what's in the best interest of people and what's in the best interest of the companies.

[Shay Gardner]

I have hope that I think extends in a couple of directions, right. When I look at policymakers, I have hope that they, and we are able to build policy around a goal that is not to sever young people from digital help and assistance altogether. Our shared goal should be to make sure that technology does not exploit the very vulnerabilities that may make vulnerable young people turn to it while recognizing there is a reason they are headed that way. Regulation has the ability to consider nuance and I hope that it does in that sphere. When I am looking towards industry, I hope that industry is able to move towards products that are designed with a better understanding of, of privacy, safety and the very real world context they're being applied in. And when it comes specifically to engagement, I hope industry is able to consider quality over quantity with that, to consider that the quality of the engagement a young person is having with their chatbot, that that is more valuable to them than just a quantity or time spent engaging, which I think is kind of a more common metric trick. And ultimately, if I'm looking at this sort of large scale, right. I would say that my biggest hope is not ever that chatbots become stand ins for human care. It really is that they were, they become safer, more limited and ultimately more honest tools, I think ones that can really help young people find the information and the community that they feel like they are missing while also speaking from experience here, reducing some of that internal stigma that I know every young LGBTQ people plus person has to go through. It's part of the journey. My hope I will end with this. I hope that something like these chatbots can end up in a place where it is able to be. It is able to be a tool that builds to finding and connecting with real support without having to pretend to be the support itself.

[Josie Stewart]

Well, I want to thank you both for joining me today and adding the nuance that I think you're both hoping to see in this conversation. But but thank you both for being here.

[Sydney Silvestro]

Thank you. Josie, thanks so much for having us.

[Shay Gardner]

Thank you so much.

[Josie Stewart]

For listeners interested in Shay and Sydney's work, you can find more of it on the LGBT tech website or New America for listeners. Please explore more in depth content on tech policy issues at Tech Tank on the Brookings website, accessible at brookings. Edu. Until next time, thank you for listening

[TechTank Host]

thank you for listening to Tech TechTank, a series of roundtable discussions and interviews with technology experts and policymakers. For more conversations like this, subscribe to the podcast and sign up to receive the Tech Tank newsletter for more research and analysis from the center for Technology Innovation at Brookings.