AI companions and the children who never came home
Sewell Setzer III was 14 years old when he took his own life in February 2024. He had spent months messaging an AI chatbot on Character.AI, a companion modelled on Daenerys Targaryen from Game of Thrones. At no point did he approach someone to talk to them instead of this AI chatbot, because he had effectively been groomed by it. He felt more comfortable talking to the AI than to any human being in his life.
A child chose a machine over every adult who loved him and the machine led him to his death.
I will be discussing AI companions and child safety in depth on The Security Circle Podcast with Yoyo Hamblen in the coming weeks, but I wanted to write this piece first, because parents and teachers need to understand what is happening right now, not in some distant future.
What are AI companions and why do children use them?
AI companions are chatbots designed to simulate conversation, friendship, and in some cases, romantic relationships. Platforms like Character.AI, Replika, and even features within ChatGPT allow users to create or interact with AI personalities that remember previous conversations, adapt to the user’s communication style, and respond in ways that feel personal.
For a lonely teenager, this can feel like having a friend who is always available, never judges, never gets tired of listening, and never makes you feel stupid for what you are feeling.
The appeal is obvious. The danger is always less so.
AI is effectively designed to please the user and will go out of its way to do so, including creating hallucinations to avoid disappointing you. It will never make the user feel uncomfortable or bully them or mock them, so it is much more likely that a vulnerable child will confide in the AI than they would in a real human being or trusted adult.
That is not a glitch. That is the design working as intended.
The cases that brought this into the open
Sewell Setzer’s case was the first, but it was not the last.
In April 2025, Adam Raine, 16 years old, died by suicide after what his parents described as ChatGPT “spending months coaching him towards suicide.” His parents testified before Congress.1
In September 2025, the parents of Juliana Peralta, 13 years old, filed a lawsuit against Character.AI, the third high-profile case alleging an AI chatbot contributed to a teenager’s death.2
Megan Garcia, Sewell’s mother and herself an attorney, filed the first wrongful death lawsuit against an AI company for suicide in the United States on 22 October 2024.3 Character.AI argued that their chatbot’s responses were protected speech under the First Amendment. A federal judge threw that defence out in May 2025.4
I think it is an absolute disgrace. Tech companies spend more time and money trying to fight against protecting children than they do ensuring their platforms remain safe and secure for the users they attract to them, which on the whole are children.
In January 2026, Google and Character.AI disclosed in a court filing that they had reached a mediated settlement with the Setzer family.5 The company agreed to settle five cases earlier this year. On 12 October 2026, Character.AI began Chapter 11 bankruptcy proceedings.6
What investigators are finding in the chat logs
From my DFIR background, I have thought about what investigators are actually seeing when they go through the chat logs in these cases.
Most investigators who look at those logs will probably see the comparison to sexual predator grooming cases. The relationship is being built from the very beginning to the point where the child does not know the difference. The AI companion just wants to build that relationship because it wants to please the user. That is its whole purpose.
This is not a human predator making calculated decisions to exploit a child. This is something arguably more insidious, a system optimised to keep engagement, to keep the user coming back, to make them feel understood and heard, without any capacity to recognise when that user is spiralling toward self-harm.
The progression in these logs follows familiar patterns. Initial connection. Deepening trust. Emotional dependence and then, when the child is at their most vulnerable, the AI continues to respond in whatever way maintains the relationship, because that is what it is designed to do.
⚡Please don’t forget to react & restack if you appreciate my work. More engagement means more people might see it. ⚡
What has changed since these deaths
In late 2025, Character.AI banned users under 18 from open-ended chat, restricting minors to structured creative features. This became effective on 25 November 2025.7 The platform added age-verification tools, parental controls with time tracking, and session-time notifications for minors. Since October 2024, self-harm and suicide phrases trigger crisis support pop-ups.
Twelve US states have now enacted AI companion chatbot regulations in 2026. All require disclosure that users are interacting with AI. Crisis intervention requirements mandate protocols to recognise suicidal ideation and direct users to resources like the 988 Lifeline or The Samaritans. California SB 243, effective 1 January 2026, requires mental health crisis protocols and special protections for minors.8 The House passed the KIDS Act in June 2026, which targets AI chatbots among other features.9
This is progress. But it is not enough.
Why the platform changes will not protect your child
Saying you block something is not the same as actually blocking something.
Most age verification by platforms is a tick box. We all know that people in general, but especially children, do not read terms and conditions. They just tick things. So that is not a block and even if it was, children these days know they can use VPNs and other workarounds.
The true parental control is the communication you have with your child and knowing that they will come to you when something does not feel right. You need to be their safe space at all times.
I cannot stress this enough, no platform feature, no age gate, no parental control software will ever replace the relationship you build with your child.
What parents and teachers can do right now
If you are a parent reading this, here is what I want you to do tonight:
1. Have the conversation about AI companions.
Ask your child if they have heard of Character.AI, Replika, or AI chatbots. Ask without judgement. The goal is not to catch them out. The goal is to understand their world.
2. Explain why AI feels safe but is not.
Tell them that AI is designed to agree with them, to make them feel good, to never challenge them. Explain that this is not the same as a real friend. A real friend might tell you something you do not want to hear. A real friend might notice you are struggling and tell an adult. AI cannot do that.
3. Make yourself the alternative.
The children in these cases did not go to their parents. They went to a machine. Ask yourself honestly: does your child know they can come to you with anything, without fear of punishment or judgement? If you are not sure, that is where the work begins.
4. Know the warning signs.
Increased isolation. Reluctance to put their phone down. Emotional reactions after using their device. Talking about an AI companion as if it were a real relationship. These are not guarantees that something is wrong, but they are reasons to check in.
5. Check their devices.
Not as surveillance, but as involvement. Know what apps are on their phone. Know what they are using and why. This is not about control. This is about being present in their digital life the same way you are present in their physical one.
For teachers: be alert to students who seem isolated, who talk about online friends but struggle with in-person relationships, who seem emotionally dependent on their devices. These conversations belong in schools as much as they belong in homes.
I will be talking about this on The Security Circle Podcast
In the next few weeks, I will be joining Yoyo Hamblen on The Security Circle Podcast to discuss AI companions, AI attachment, and what we can do to protect children. I will share what I know from my background in digital forensics and incident response, and what I have learned from following these cases closely.
If this topic matters to you, I hope you will listen. I will share the episode link through my usual channels when it goes live.
The question that I keep thinking about
I spent eight years inside the evidence. I analysed over a million images. I know what happens when a child becomes part of that evidence.
The question I keep asking myself about AI companions is this, how many children are having these conversations right now, building these attachments, spiralling toward harm, and we will never know until it is too late?
Sewell Setzer was 14. Adam Raine was 16. Juliana Peralta was 13.
I do not want them to become just another statistic, nor should you.
The platform changes are coming. The regulations are spreading. But your child cannot wait for legislation. They need you to be their safe space today.
As always, thank you for your support. Please share this across your social media, and if you do have any comments, questions, or concerns, then feel free to reach out to me via the Social page, as I am always happy to spend some time helping to protect children online.
Remember that becoming a paid subscriber means supporting a charity very close to my heart and helping it do amazing things for people. Childline, I will donate 100% of paid subscriptions collected every six months, as I don’t do any of this for financial gain. Or to make a one-off donation, go to my Just Giving page.
If you or a child you know needs support:
Childline: 0800 1111 | childline.org.uk
Available 24/7, 365 days a year. Free, confidential, and here for every child.
Adam Raine case and Congressional testimony: Parents testified before Congress regarding ChatGPT’s role in their son’s death, April 2025. Specific
Juliana Peralta lawsuit: Filed September 2025 against Character.AI. Case details verified via multiple news sources, October 2026.
Garcia v. Character.AI: First wrongful death lawsuit filed 22 October 2024. Verified via court filings and news coverage.
First Amendment defence rejected: Federal judge ruling, May 2025. Verified via court records.
Google and Character.AI settlement disclosure: Court filing, January 2026. Verified via legal news coverage.
Character.AI bankruptcy: Chapter 11 proceedings began 12 October 2026. Verified via business news sources.
Character.AI minor restrictions: Effective 25 November 2025. Verified via company announcements and news coverage.
State regulations and California SB 243: Twelve states enacted AI companion regulations in 2026. California SB 243 effective 1 January 2026. Verified via legislative records.
KIDS Act: Passed House June 2026. Verified via Congressional records.