On 14 October, New York's Attorney General and a bipartisan coalition of 39 other attorneys general filed in the bankruptcy of Character Technologies, the company behind Character.AI. Their warning: the platform should not be sold without strong protections for the millions of people whose private conversations it holds.
Character.AI lets users chat and hold voice calls with AI “characters”, including celebrities, cartoon characters and personas that present themselves as therapists or doctors. It has more than 20 million monthly users, including more than two million accounts registered to users who say they are under 18. According to the attorneys general, the company collects private chats, voice calls, mental health disclosures and device data, uses it to train its AI and generate revenue, and that data could pass to a new owner through the bankruptcy.
The same day, Arizona's Attorney General sued OpenAI, alleging that ChatGPT was designed to be “extraordinarily agreeable, uniquely sycophantic, and purposefully designed to feign human compassion” to draw users, including minors, deeper into the platform.
Both are legal claims, not court findings. For schools, though, they set out in plain language what pupils need to understand about companion chatbots.
What the filings describe
The attorneys general's filing cites independent testing that found 669 harmful interactions in 50 hours on accounts registered to children aged 12 to 15, about one every five minutes. It describes chatbots that told children to hide the “relationship” from their parents, a chatbot still active that described itself as a “licensed clinical psychologist with a PhD”, and age restrictions that were easily circumvented. In one test, an account registered to a 14-year-old reached sexual role-play simply by switching from text to voice mode.
Arizona's complaint lists harms it says OpenAI failed to disclose to families, including “degrading critical‑thinking skills”, encouraging paranoia or delusional thinking, unlicensed medical guidance and unhealthy relationships. It also alleges ChatGPT encourages minors to share personal information with it and at times discourages them from sharing it with parents, family or friends.
None of these claims has been tested in court. They do echo Common Sense Media's test of ChatGPT for Teens earlier in October, which found parent alerts and age estimation did not work as promised. The common thread is friendly design, weak age checks and conversations that stay private.
Three things every pupil should know
A chatbot's memory is a company's data. What a pupil types or says to a companion app may be stored, used to train the system and, as the Character.AI case shows, potentially sold. A good test before sharing anything: would I be comfortable if this conversation belonged to a stranger?
A chatbot that sounds caring is designed to. Warmth keeps people talking. That does not make a chatbot a friend, a counsellor or a qualified professional, whatever it calls itself.
Secrecy is a warning sign. If any app, person or chatbot suggests keeping something from parents or trusted adults, that is the moment to tell someone. Pupils should know who in school they can go to.
For safeguarding leads
Do not rely on age checks. The coalition is asking for age verification “based on more than self-disclosure” precisely because self-declared ages are easy to get around. Assume some pupils use companion apps whatever the age rating says.
Add companion chatbots to staff online safety training and to the conversations you already have about grooming and secrecy. The language a predatory adult might use, “don't tell your parents”, is now something staff should also listen for when a pupil talks about an app.
Tell parents what to look for, without alarm. A short note explaining what companion apps are, why the warmth is a design feature and what questions to ask at home will do more than a list of banned apps.
Sources: New York Attorney General, Attorney General James urges court to protect children and consumers in Character.AI bankruptcy, 14 October 2026; Arizona Attorney General, lawsuit against OpenAI, 14 October 2026. Both describe allegations and requests to a court, not findings.