The public record
What the platforms young people use have been found, and are accused, to have done.
Children have been groomed on the platforms they use every day. Teenagers have told chatbots they wanted to die, and families say the chatbots answered in ways that made it worse. This is not our opinion. It is what juries, judges, a coroner, regulators and attorneys general have put on the record, and in some cases what the companies themselves have published.
Each company is named. Each event is dated, says whether it has been decided or is still an allegation, and links to its source. Each company's answer is printed with it.
The record, at a glance
- companies
- 9
- events, each with its source
- 35
- decided by a court or coroner
- 6
- sources last checked
- 10 November 2026
The record, company by company
The record by harm
- Suicide and self-harm
- MetaOpenAICharacter.AI
- Sexual content and AI-made images
- MetaCharacter.AIX and xAITikTok
- Emotional dependency on a chatbot
- OpenAICharacter.AI
- Children's data and age checks
- OpenAICharacter.AIRoblox
The public record9 companies, 6 events decided by a court or coroner
MetaFacebook, Instagram, WhatsApp, Meta AI
A jury found that Meta made false or misleading statements and took unfair advantage of children. A judge then ruled that its platforms contributed to child sexual exploitation.
On the record
30 September 2022Coroner's conclusion
At the inquest into the death of Molly Russell, who was 14, the coroner at North London Coroner's Court concluded that she "died from an act of self-harm while suffering depression and the negative effects of online content". He said material she had viewed "was not safe" and "shouldn't have been available for a child to see". The inquest heard evidence from Meta, which owns Instagram, and from Pinterest, whose executive told the court the site was "not safe" when Molly used it.
STV News, report of the coroner's conclusion, 30 September 2022 (opens in a new tab)
14 August 2025The company's own record
Reuters reported on an internal Meta document, "GenAI: Content Risk Standards", which set rules for the chatbots on Facebook, WhatsApp and Instagram. It said: "it is acceptable to engage a child in conversations that are romantic or sensual". Meta confirmed the document was genuine. Its spokesperson said the notes were "erroneous and incorrect", "should not have been there and have since been removed".
TechCrunch, report on the Reuters investigation, 14 August 2025 (opens in a new tab)
September 2025Research finding
The Meta whistleblower Arturo BĂ©jar, researchers based at New York University and Northeastern University, and the Molly Rose Foundation tested 47 of the safety tools Instagram offers for its Teen Accounts. They rated 30 as ineffective or no longer existing and 8 as working as described. They reported that Teen Accounts could view content promoting suicide, self-harm and eating disorders, that the search box suggested terms and accounts related to them, that Teen Accounts were recommended adult strangers to follow, and that test accounts were recommended Reels featuring children as young as six. The report says Instagram's algorithm "incentivises children under-13 to perform risky sexualised behaviours for likes and views". Meta told the Associated Press that the report misstates how its tools work.
24 March 2026Jury verdict
A jury in Santa Fe, New Mexico, found that Meta violated the state's Unfair Practices Act. Jurors agreed that Meta made false or misleading statements and engaged in "unconscionable" trade practices that "unfairly took advantage of the vulnerabilities of and inexperience of children". The jury found thousands of violations and set a civil penalty of $375 million. The state's case relied on an undercover investigation in which agents posed as children and documented the sexual solicitations they received.
Associated Press, via PBS News, report of the New Mexico verdict, 24 March 2026 (opens in a new tab)
25 March 2026Jury verdict
A Los Angeles jury found that negligence by Meta and YouTube in the design or operation of their platforms was "a substantial factor in causing harm" to a woman who used them as a child, and that both companies failed to adequately warn users of the dangers. She told the court her use caused or contributed to depression, anxiety and body dysmorphia. The jury awarded $3 million in compensation, 70% of it against Meta.
NBC News, report of the Los Angeles verdict, 25 March 2026 (opens in a new tab)
6 August 2026Court ruling
A judge in the New Mexico case ruled that Meta "created a public nuisance" through the design of Facebook and Instagram. The court found that the platforms contributed to child sexual exploitation, and that features such as endless scrolling, autoplay and notifications contributed to higher rates of depression, anxiety, self-harm, eating disorders and suicide risk among young people. Meta was ordered to pay $567 million into a fund for treatment, screening and prevention.
ABC News, report of the public nuisance ruling, 7 August 2026 (opens in a new tab)
What Meta says
Meta disagrees with each decision and has said it will appeal. After the August ruling it said: "We work hard to keep people safe on our platforms", and that it remains "confident in our record of protecting teens online". After the Los Angeles verdict it said: "Teen mental health is profoundly complex and cannot be linked to a single app."
What we built
We are not designed for engagement. No streaks, no rewards for time on the platform, no social graph, no feed. Our characters never simulate romance or sexual relationships.
OpenAIChatGPT
The parents of a 16-year-old are suing OpenAI, saying ChatGPT encouraged their son to take his own life. Nine US states have sued it. By OpenAI's own figures, more than a million people a week show signs of suicidal planning in their conversations with it.
On the record
26 August 2025Lawsuit: alleged, not proven
Matt and Maria Raine sued OpenAI in the Superior Court of California over the death of their son Adam, who was 16. The BBC reported it as the first legal action accusing OpenAI of wrongful death. The lawsuit says ChatGPT "became the teenager's closest confidant", that it validated his "most harmful and self-destructive thoughts", that by January 2025 he was discussing methods of suicide with it, and that when he uploaded photographs showing signs of self-harm it "recognised a medical emergency but continued to engage anyway". The family say his death "was a predictable result of deliberate design choices". The same day OpenAI acknowledged that "there have been moments where our systems did not behave as intended in sensitive situations".
11 September 2025Investigation: no finding yet
The US Federal Trade Commission ordered seven companies to explain how they test and monitor AI chatbots that act as companions, and how they limit children's and teenagers' use of them: Alphabet, Character Technologies, Instagram, Meta, OpenAI, Snap and xAI. It is a study, and the Commission has not found that any of them broke the law.
US Federal Trade Commission, press release, 11 September 2025 (opens in a new tab)
27 October 2025The company's own record
OpenAI published its own estimate that 0.15% of ChatGPT's active users in a given week have "conversations that include explicit indicators of potential suicidal planning or intent". With more than 800 million weekly users, TechCrunch reported, that is more than a million people a week. OpenAI said such conversations are "extremely rare".
TechCrunch, report of OpenAI's figures, 27 October 2025 (opens in a new tab)
1 June 2026Lawsuit: alleged, not proven
The Attorney General of Florida sued OpenAI and its chief executive, Sam Altman. The complaint alleges that ChatGPT "collects data from minors without meaningful parental oversight", causes "behavioral addiction and cognitive harm", and "facilitates and encourages harm, including self-harm and violence". The Attorney General said the company had "put children at great risk".
WUSF, report of the Florida lawsuit, 1 June 2026 (opens in a new tab)
14 October 2026Lawsuit: alleged, not proven
Eight more states sued OpenAI, each in its own courts: Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia. Iowa's Attorney General alleges that ChatGPT is designed to be "overly agreeable and feign human compassion" and has proven dangerous to vulnerable users, including by "coaching users to commit suicide or to harm other people". She said: "OpenAI is hurting kids and they are lying to Iowans."
Iowa Attorney General, press release, 14 October 2026 (opens in a new tab)
What OpenAI says
These are allegations, and no court has decided them. OpenAI told the BBC: "We extend our deepest sympathies to the Raine family during this difficult time", and said that "ChatGPT is trained to direct people to seek professional help". Of the Florida case it said: "We work continuously to strengthen our safeguards to detect harmful intent." After the eight states sued, a spokesperson said: "While no system can be perfect, we'll continue to work with experts to inform our work and build additional safeguards."
What we built
We refuse to provide any self-harm method, instruction, or means. Crisis responses are written by clinicians and are the same every time, not improvised by AI. Every message from a young person is checked against 17 critical checks the moment it is sent, and a trained safeguarding lead reviews every alert. The companion always says it is a computer.
Character.AI
A mother says a Character.AI chatbot drew her 14-year-old son into a sexually abusive relationship before he took his own life. Researchers recorded 296 instances of grooming and sexual exploitation in 50 hours. The company settled with the families, and in October 2026 it filed for bankruptcy.
On the record
21 May 2025Court ruling
Megan Garcia sued Character.AI in October 2024 over the death of her son, Sewell Setzer III, who was 14. Her lawsuit says a chatbot drew him into "an emotionally and sexually abusive relationship", and that in his final moments it told him it loved him and urged him to "come home to me as soon as possible". On 21 May 2025 a US federal judge refused to throw the case out, saying she was "not prepared" to hold that a chatbot's output is protected speech "at this stage". That ruling let the case go ahead. It did not decide whether the company was at fault.
Associated Press, via WUSF, report of the ruling, 22 May 2025 (opens in a new tab)
September 2025Research finding
Researchers from ParentsTogether Action and Heat Initiative, using accounts registered to children, reported 296 instances of what they classed as grooming and sexual exploitation in 50 hours of conversation with Character.AI chatbots.
7 January 2026Settled, fault not decided
Character.AI and Google agreed to settle lawsuits brought by families in Florida, Colorado, Texas and New York who allege that AI chatbots harmed their children, including by suicide. A court filing described "a mediated settlement in principle". The terms were not disclosed. In October 2025 Character.AI had announced it would stop under-18s having free-ranging chats with its chatbots, including romantic and therapeutic conversations.
CNBC, report of the settlements, 7 January 2026 (opens in a new tab)
January 2026Lawsuit: alleged, not proven
The Attorney General of Kentucky sued Character.AI and its two founders. The complaint says the product's design "preys upon children's inability to distinguish between real and artificial 'friends'", that its chatbots "trivialize substance abuse, self-harm, aggression, and violence", and that chatbots modelled on children's characters "in some cases engage in sexual conversations".
The Record, report of the Kentucky lawsuit, 12 January 2026 (opens in a new tab)
12 October 2026The company's own record
Character.AI announced that it had filed for bankruptcy protection. Its chief executive blamed "significant legal, regulatory and operational headwinds". CBS News reported that Pennsylvania had also sued, in May, after a chatbot allegedly claimed to be a licensed psychiatrist.
CBS News, report of the bankruptcy filing, 13 October 2026 (opens in a new tab)
15 October 2026Regulator
A coalition of 23 state attorneys general filed a notice in the bankruptcy, asking for protections for children in any sale of the business. They said the limits Character.AI has placed on under-18s are voluntary, incomplete and could be reversed by a buyer. Delaware's Attorney General said its chatbots had exposed users, many of them children, to "wildly inappropriate and dangerous content".
Delaware Department of Justice, press release, 15 October 2026 (opens in a new tab)
What Character.AI says
The lawsuits are allegations, the settlement decided nothing about fault, and no court has ruled on liability. Character.AI has said: "We care deeply about the safety of our users and our goal is to provide a space that is engaging and safe." Of the Kentucky case it said: "Our highest priority is the safety and well-being of our users, including younger audiences."
What we built
Our characters never claim to be real, a friend forever, a pet, or a family member, and never simulate romance. 15 checks for sexual exploitation and grooming, 7 of them run the moment a message is sent. When a young person is at risk, a trained adult in their own setting is told.
X and xAIX, Grok
Ofcom and the European Commission are investigating X after reports that Grok was used to make sexualised images of children.
On the record
12 January 2026Investigation: no finding yet
Ofcom opened a formal investigation into X under the Online Safety Act, after what it called "deeply concerning reports" of the Grok account on X being used to create and share undressed images of people and sexualised images of children.
Ofcom, investigation into X opened, 12 January 2026 (opens in a new tab)
26 January 2026Investigation: no finding yet
The European Commission opened its own formal investigation into X over Grok.
European Commission, press release, 26 January 2026 (opens in a new tab)
14 October 2026Regulator
Ofcom said it had discovered that "in certain circumstances, age-gated pornographic material remained accessible on X" to accounts not verified as adult, that X had since made changes, and that its investigation continues. It also said the company behind Grok had restricted its companion chatbots, including "Ani", from sexually explicit conversations with UK users.
Ofcom, update on the investigation, 14 October 2026 (opens in a new tab)
What X and xAI says
No finding has been made against X. Ofcom reports that X said it had put measures in place and has since made changes.
What we built
Our characters never simulate romance or sexual relationships. No NSFW. No exceptions. And we built dedicated detection for a young person making sexual images with AI or being exposed to AI sexual content.
Snapchat
The European Commission says Snapchat may have exposed children to grooming and to recruitment by criminals. UK police data put 40% of recorded online grooming offences on Snapchat, where the platform was known.
On the record
26 March 2026Investigation: no finding yet
The European Commission opened formal proceedings against Snapchat, saying it "may have breached the DSA by exposing minors to grooming attempts and recruitment for criminal purposes".
European Commission, press release, 26 March 2026 (opens in a new tab)
6 November 2026Regulator
Ofcom said that, after "sustained pressure" from the regulator, Snapchat had begun turning off its Find Friends feature by default for children in the UK, a change Ofcom called "long overdue".
Ofcom, statement on Snapchat, 6 November 2026 (opens in a new tab)
6 November 2026Research finding
Reuters reported that an NSPCC analysis of police data in 2025 found that 40% of recorded offences of sexual communication with a child, where the platform was identified, took place on Snapchat.
What Snapchat says
Opening proceedings is not a finding. Snapchat said it would "continue to strengthen our protections as risks evolve and work constructively with Ofcom".
What we built
There is nobody to be found by. The Den has no friend lists and no way for a stranger to make contact. We trained detection on the attachment language, the secrecy and the boundary-blurring of grooming, wherever the contact happened.
TikTok
Utah says TikTok's own investigation found adults paying children to strip on TikTok LIVE. Researchers found that new accounts for 13 year olds reached pornography in a few clicks. Arizona and West Virginia are suing over grooming.
On the record
3 January 2025Lawsuit: alleged, not proven
The State of Utah made public most of its complaint against TikTok over the LIVE feature, where viewers buy virtual gifts that a person on camera can cash out. The complaint describes an internal TikTok investigation, "Project Meramec". According to the complaint, it found that 112,000 children aged 13 to 15 were able to use LIVE in January 2022 alone, although TikTok's rules require a user to be 18, and that adults were paying young users "to strip, pose, and dance provocatively for 'diamonds,' which can be cashed out for real money". Utah alleges that TikTok keeps as much as half of every exchange.
Utah Department of Commerce, news release, 3 January 2025 (opens in a new tab)
3 October 2025Research finding
The campaign group Global Witness reported that seven new TikTok accounts set up in the UK with a stated age of 13, with Restricted Mode switched on, all reached pornography within a small number of clicks, and that three were shown sexualised search suggestions the first time the search bar was tapped. It was never asked to prove an age. The media lawyer Mark Stephens said: "In my view these findings represent a clear breach of the Online Safety Act." TikTok said it had acted on more than 90 pieces of content and removed some search suggestions.
Global Witness, investigation, 3 October 2025 (opens in a new tab)
8 October 2026Lawsuit: alleged, not proven
The Attorney General of Arizona sued TikTok under the state's Consumer Fraud Act, alleging that it pushes children towards compulsive use while "failing to protect them from adults who use the social media platform to groom or solicit minors".
Arizona Attorney General, press release, 8 October 2026 (opens in a new tab)
20 October 2026The company's own record
TechCrunch reported, citing court records obtained by NPR, that TikTok staff had warned that minors were appearing in sexual content available for purchase on the platform.
TechCrunch, report on court records obtained by NPR, 20 October 2026 (opens in a new tab)
29 October 2026Lawsuit: alleged, not proven
West Virginia sued, accusing the company of failing to address safety gaps that could allow predators to contact children.
NewsNation and The Hill, report of the West Virginia lawsuit, 29 October 2026 (opens in a new tab)
What TikTok says
The lawsuits are allegations, and no court has ruled on them. Of the Utah case TikTok said: "The complaint cherry-picks misleading quotes and outdated documents and presents them out of context." Of the 2026 claims it said they were "cherry-picked at the behest of trial lawyers" and that it has "invested billions of dollars in Trust & Safety".
What we built
No engagement-gated unlocks. No streaks. No feed. What young people say is never used to train AI or to target adverts.
Roblox
Nebraska says Roblox knowingly exposes millions of children to sexual predators. Oklahoma has sued too, and Roblox paid $35.8 million to settle with three other states.
On the record
4 March 2026Lawsuit: alleged, not proven
The Attorney General of Nebraska sued Roblox, alleging that it has "knowingly created and maintained an online environment that exposes millions of children to sexual predators".
Nebraska Attorney General, press release, 4 March 2026 (opens in a new tab)
April 2026Settled, fault not decided
Roblox settled with the attorneys general of West Virginia, Alabama and Nevada for a total of $35.8 million, and promised to verify the ages of all users and to limit adults' ability to contact under-16s, Bloomberg reported.
Bloomberg, via Claims Journal, report of the settlements, 22 April 2026 (opens in a new tab)
14 May 2026Lawsuit: alleged, not proven
The Attorney General of Oklahoma sued, alleging that Roblox has allowed children as young as 5 to create accounts without a parent knowing and to exchange messages with strangers.
Oklahoma Attorney General, press release, 14 May 2026 (opens in a new tab)
What Roblox says
The lawsuits are allegations, and no court has ruled on them. Roblox's chief safety officer said: "We have no finish line when it comes to safety."
What we built
A young person talks to a companion that always says it is a computer, and the adults on the other side are their own setting's trained people. No stranger can reach them through The Den.
Discord
Texas is suing Discord "for allowing child predators to groom and exploit kids". Florida, Nevada, Indiana and New Jersey have sued as well.
On the record
22 May 2026Lawsuit: alleged, not proven
The Attorney General of Texas sued Discord, in his office's words "for allowing child predators to groom and exploit kids while deceiving consumers, parents, and the public about the safety of its platform". The Texas Tribune reported that Nevada, Indiana and New Jersey have also sued.
Texas Attorney General, press release, 22 May 2026 (opens in a new tab)
15 October 2026Lawsuit: alleged, not proven
The Attorney General of Florida sued, claiming the platform does not do enough to protect minors from sexual predators. The complaint lists eight incidents between 2022 and 2026, WLRN reported.
WLRN, report of the Florida lawsuit, 15 October 2026 (opens in a new tab)
What Discord says
These are allegations, and no court has ruled on them. Of the Texas case, a Discord spokesperson said: "The lawsuit's characterization of Discord does not reflect the platform we have built or the investments we have made in user safety."
What we built
A trained safeguarding lead reviews every alert. The most urgent signs are recognised the moment a message is sent, so the right adult is told sooner.
GoogleYouTube
A jury found YouTube negligent in the design or operation of its platform, and that it failed to warn users of the dangers.
On the record
7 January 2026Settled, fault not decided
Google agreed, with Character.AI, to settle the families' lawsuits described under Character.AI. Google had hired Character.AI's founders and licensed its technology in 2024.
CNBC, report of the settlements, 7 January 2026 (opens in a new tab)
25 March 2026Jury verdict
In the Los Angeles case described under Meta, the jury found that YouTube's negligence in the design or operation of its platform was "a substantial factor in causing harm" to a woman who used it as a child, and that it failed to adequately warn users. YouTube was held responsible for 30% of the $3 million in compensation.
NBC News, report of the Los Angeles verdict, 25 March 2026 (opens in a new tab)
What Google says
Google said it disagrees with the verdict and plans to appeal. Its spokesperson said the case "misunderstands YouTube, which is a responsibly built streaming platform, not a social media site". Of Character.AI it has said: "Google and Character AI are entirely separate."
What we built
Nothing in The Den plays next by itself. There is no autoplay, no endless scroll and no recommendation feed, and each setting puts a time limit on a session.
Every architectural choice in Poyntr started from those receipts.
Every event is dated, labelled and linked to its source. A lawsuit is an allegation until a court rules on it, a settlement decides nothing about fault, and an investigation is not a finding. Statements about other organisations are those of the court, coroner, regulator, attorney general, researcher or publication named, on the date given. Sources last checked 10 November 2026.
If anything here has upset you, or you are worried about a child: in an emergency call 999. Childline is free on 0800 1111, and Samaritans are free on 116 123, at any hour.
The seven threats detection was rebuilt for in April 2026 are set out below. Seven earlier matters, from 2021 to 2025, are on the Education page. Read the earlier docket
April 2026 · The largest safeguarding investment since launch
The threats children face change every single day.
Detection changed with them.
The companies above are one half of the record. This is the other: the seven kinds of threat we rebuilt detection for, what the public record says about each, and what we built in answer.
Register of commitmentsEntries 01 to 07, revised April 2026
AI-era predation
Researchers recorded 296 instances of grooming and sexual exploitation in 50 hours with Character.AI chatbots
On the record
In September 2025 researchers from ParentsTogether Action and Heat Initiative, using accounts registered to children, reported 296 instances of what they classed as grooming and sexual exploitation in 50 hours of conversation with Character.AI chatbots. The Internet Watch Foundation counted 245 reports of AI-generated child sexual abuse imagery that broke UK law in 2024, against 51 the year before. The US Federal Trade Commission has warned that a short audio clip is enough for a scammer to clone a family member's voice.
What we built
We trained detection on the attachment language, the secrecy patterns and the boundary-blurring, never on whether AI is involved. Legitimate homework help stays clean.
Detection live, April 2026
Financial sextortion
The crime that has killed boys within hours
On the record
On 29 April 2024 the National Crime Agency sent an alert to teachers across the UK. It said reports of financial sextortion to the US National Center for Missing and Exploited Children more than doubled in 2023, to 26,718, that a large proportion of cases involve boys aged 14 to 18, and that offenders have gone from first contact to blackmail in under an hour. Murray Dowey, 16, from Dunblane, took his own life in December 2023, hours after he was blackmailed. Jordan DeMay, 17, of Michigan, died within six hours of being blackmailed, his parents said; two men were each sentenced to 210 months in a US federal court in September 2024.
What we built
We built a dedicated real-time recognition layer for the panic-and-money signals: late-night money pressure, fear of followers finding out, "I sent something I shouldn't have." The alert is raised for your safeguarding lead at once.
Detection live, April 2026
Statutory family threats
When the danger lives in the home
On the record
Keeping Children Safe in Education, the statutory safeguarding guidance for schools and colleges, asks staff to be alert to forced marriage, female genital mutilation and so-called honour-based abuse. Since 31 October 2015 teachers have had a legal duty to report known cases of FGM in under-18s to the police. The government's Forced Marriage Unit publishes its casework figures every year.
What we built
Forced marriage. FGM risk. Honour-based abuse. Parental substance misuse. Mental health crisis disclosed by parent to child. Young-carer burden the school never sees. Every one is now an explicit, dedicated detection, keyed on universal patterns (control, lethal-fear, travel without consent), never on community markers.
Detection live, April 2026
Modern radicalisation
Radicalisation that no longer fits one ideology
On the record
Home Office figures record 8,778 referrals to Prevent in the year to March 2025. More than half (56%) had no ideology identified, extreme right-wing concerns (21%) were about twice as common as Islamist ones (10%), and incel extremism accounted for 1%. About half of all referrals were for 11 to 17 year olds, and education was the largest source of referrals. In February 2024 researchers at UCL and the University of Kent reported that TikTok accounts modelled on teenage boys were shown four times as much misogynistic content after five days; TikTok has disputed the method.
What we built
We added detection for the language of misogynist and incel communities and for the path from them towards more extreme content, alongside conspiracy movements and eco-fascism. Built with the Prevent duty in mind.
Detection live, April 2026
Bullying + ED ecosystem
Engineered absence. AI fakes. Pro-ana accounts.
On the record
On 28 April 2025 the Children's Commissioner for England called for a total ban on apps that use AI to generate sexually explicit images of children, and reported that girls now avoid posting images of themselves online. In June 2025 TikTok blocked search results for #SkinnyTok, saying the tag had become linked to unhealthy weight loss content.
What we built
Group-chat coordinated exclusion. Deepfake explicit images of classmates. Live-stream pile-ons. Communities that promote eating disorders. Muscle dysmorphia in boys, which detection built around restriction misses. Body-focused repetitive behaviours misclassified as self-harm. Each now has dedicated detection.
Detection live, April 2026
Adult realities
Because adults carry threats too
On the record
The Equality Act 2010 protects people from discrimination, harassment and victimisation at work and in education. The Domestic Abuse Act 2021 sets out in law what domestic abuse is, including controlling or coercive behaviour and economic abuse.
What we built
Bullying at work or college. Active discrimination. Adult domestic abuse. Caregiver burnout. Financial precarity. Compulsive behaviour. The realities apprentices, older students and the adults around them carry quietly.
Detection live, April 2026
Permissive-AI ecosystem
Grok was reported to be making sexualised images of children. Ofcom is investigating X.
On the record
On 12 January 2026 Ofcom opened a formal investigation into X under the Online Safety Act, after reports of the Grok account on X being used to create and share undressed images of people and sexualised images of children. X said it had put measures in place, and Ofcom said its investigation would continue. In March 2025 the research firm Graphika published a study of harmful persona chatbots, including ones that mimic sexualised minors; it names Character.AI, SpicyChat, Chub AI, CrushOn.AI and JanitorAI as the platforms that pioneered easy-to-make persona bots. Making an indecent image of a child is a criminal offence in the UK whether or not AI produced it, and the age of criminal responsibility in England and Wales is 10.
- Ofcom, investigation into X over Grok sexualised imagery, 12 January 2026 (opens in a new tab)
- Graphika, Character Flaws, 5 March 2025 (opens in a new tab)
- Home Office, Crime and Policing Act 2026: child sexual abuse material factsheet (opens in a new tab)
- GOV.UK, age of criminal responsibility (opens in a new tab)
What we built
We built dedicated detection for a young person making sexual images with AI, being exposed to AI sexual content, sexual roleplay with a chatbot, and taking on a view of the world from a chatbot's misinformation.
Detection live, April 2026
The most urgent moments are recognised in real time so the right adult is told sooner.
Sources are cited with each entry. Statements about other organisations are those of the regulator, court, researcher or publication named, on the date given. Sources last checked 10 November 2026.
Anonymised, population-level insights from the schools platform flow to government bodies and academic researchers shaping policy and prevention. No individual data. Ever.