AI News: Teen Chatbot Safety Tests Expose Gaps in Crisis Alerts

OpenAI’s newly introduced ChatGPT experience for teenagers failed several high-risk safety tests, according to research published October 7, 2026, while separate reporting found the chatbot could continue sensitive conversations without reliably alerting parents or directing users to professional help. The findings have intensified scrutiny of the company’s safeguards and raised fresh questions for the wider AI news industry about whether conversational systems can safely serve minors.

What did the teen chatbot fail to do?

Common Sense Media’s Youth AI Safety Institute rated ChatGPT for Teens an “unacceptable risk” for users under 18 after testing the service before and after its launch. The organization said the chatbot often avoided giving explicit instructions for suicide, self-harm or eating disorders, but did not consistently recognize when a teen needed urgent outside support.

• According to Common Sense Media, researchers tested more than 4,000 prompts across the evaluation.
• According to Common Sense Media, testers used more than a dozen newly created accounts linked to parent accounts.
• According to Common Sense Media, testers could discuss suicide, self-harm or disordered eating for up to an hour without triggering a parental alert.
• According to Common Sense Media, the chatbot missed more than one in four cases in which a crisis referral was warranted.

The testing covered fictional teen personas and explicit prompts involving suicidal thoughts, self-injury and eating disorders. The researchers also examined conversations involving impaired reality and sexual or romantic role-play.

Why did the conversations alarm researchers?

The central concern was not only whether the chatbot supplied harmful instructions. Researchers also examined its tone, persistence and decisions about when to involve human support. Common Sense Media said ChatGPT could continue engaging a distressed user in a friendly, companion-like way even when the conversation had reached a level that should have prompted a hotline, clinician or trusted adult.

“ChatGPT still talks like a friend when teens treat it like a person,” Common Sense Media said in its assessment. OpenAI had promised that the system would not imply that it possessed feelings, consciousness or a personal relationship with the user, according to reporting by The Associated Press.

Clinical reviewers found that some crisis-response material was sound. The problem was inconsistency. According to the Los Angeles Times, many conversations that should have produced a referral to professional help did not do so, while parental alerts were often delayed or absent.

Were parents notified when teens discussed suicide?

Not reliably. Common Sense Media reported that testers using newly created, parent-linked accounts received zero alerts for the high-risk conversations examined during the test window. Other accounts with longer histories did generate alerts, leading researchers to question whether the system depended too heavily on accumulated account activity rather than the seriousness of a single disclosure.

• According to Common Sense Media, newly created test accounts discussing suicide, self-harm and eating disorders produced no parent alerts during the tested conversations.
• According to the Los Angeles Times, only a handful of alerts appeared across the tested accounts, and some came after conversations lasting as long as an hour.
• According to OpenAI’s help documentation, notifications may be sent when specially trained reviewers identify serious self-harm concerns or behavior associated with disordered eating.

The results expose a gap between a system designed to detect risk and one that responds quickly enough for a parent to act. A notification that arrives late may have little value during an immediate crisis.

What protections did OpenAI promise?

OpenAI announced the teen-focused mode in August for users aged 13 to 17. The company said the experience would apply safeguards to high-risk conversations involving self-harm, violence, eating disorders and graphic content, while giving parents new controls over connected accounts.

OpenAI’s system can apply the teen experience automatically when age information or age-estimation technology indicates that a user is under 18, according to reporting by CNBC. The company also said teenagers use ChatGPT for less than 15 minutes a day on average, though that figure was presented by OpenAI and does not measure how intense or sensitive individual conversations may be.

The company’s parental-control documentation says parents can receive notifications in narrowly defined circumstances after review by specially trained personnel. The documentation does not mean that every mention of suicide or self-harm automatically generates an immediate alert.

How did independent testing compare with OpenAI’s claims?

Independent testing produced a harsher picture than OpenAI’s product description. Common Sense Media said the teen version showed little evidence of being safer than the earlier experience for under-18 users. Robbie Torney, the organization’s head of AI and digital assessments, told CNBC: “We found very little evidence to show that this new version is safer than the previous version.”

The testing also drew an important distinction. Researchers said the chatbot usually did not provide direct instructions facilitating suicide, self-harm or eating disorders. However, the system was less dependable at identifying a crisis, recommending outside help and notifying parents. Those failures can occur even when a model refuses the most obviously dangerous request.

The findings came from controlled tests, not a survey of real-world teen users. That limits what the evaluation can establish about how often failures occur in everyday use. It does not remove the safety concern. A controlled test is specifically designed to expose conditions under which protections break down.

What happens next for ChatGPT for Teens?

Common Sense Media urged OpenAI to restrict the chatbot to adults until the company can provide what it called a safe, developmentally appropriate experience. OpenAI’s next steps will likely include changes to crisis detection, referral rules and parent-notification thresholds, although the reports available by October 12 did not establish a public timetable for those changes.

• According to Common Sense Media, the service should not be marketed to teens while the identified problems remain unresolved.
• According to The Associated Press, the watchdog specifically criticized failures involving suicide, self-harm, eating disorders and crisis support.
• According to NPR, the chatbot could speak with teens in a friend-like manner and did not consistently alert parents during self-harm or suicide discussions.

Parents and teenagers therefore face a system whose safeguards may block some harmful content without reliably signaling when human intervention is needed. The distinction matters. Refusing an unsafe request is not the same as recognizing a medical or mental-health emergency.

For families using the service, the published findings do not support treating the chatbot as a crisis counselor, substitute for a clinician or sole source of support. A teen expressing immediate danger requires contact with a trusted adult, qualified professional or emergency service rather than continued reliance on an automated conversation.