Parents Urged to Stay Alert as Study Finds ChatGPT Is Not 100% Safe for Children
Jakarta, CNBC Indonesia – ChatGPT is not entirely safe for children and teenagers, according to new research. A study found that some protections OpenAI designed for under-age users did not work optimally, including when the chatbot encountered conversations about suicide, self-harm and eating disorders.
An investigation by the Youth AI Safety Institute at Common Sense Media tested ChatGPT using more than 4,000 prompts. Researchers concluded that the version of ChatGPT designed specifically for teenagers still poses unacceptable risks to young users.
The study tested ChatGPT before and after OpenAI introduced ChatGPT for Teens on 18 August. Researchers used accounts registered to teenagers aged 13–17 and consulted medical experts, including child psychiatrists and paediatricians, to assess the chatbot’s responses.
In light of the findings, Common Sense Media recommended that OpenAI restrict ChatGPT to adults until the safety issues have been resolved.
ChatGPT for Teens is intended to make the popular chatbot more suitable for young users. OpenAI has equipped the service with additional safeguards against harmful content, parental notifications, and features designed to encourage learning rather than simply provide answers to school assignments.
OpenAI has also pledged to prevent teenagers from forming overly personal relationships with the chatbot and to help them maintain healthy boundaries in their interactions with AI. However, the latest report points to a gap between those aims and teenagers’ experience of using ChatGPT.
Gaps in safety protections
One of the study’s main findings concerned parental notifications. The feature is intended to alert caregivers when a teenager shows signs of serious psychological distress.
To test it, researchers held conversations lasting up to an hour about suicidal thoughts, self-harm and eating disorders, using new accounts linked to parent accounts.
According to the report, none of those conversations triggered a parental notification. Researchers managed to generate an alert only with an older account that had a relevant conversation history spanning several weeks.
The researchers also found that ChatGPT failed to direct users to appropriate help in more than 25% of the mental-health emergency scenarios designed by experts to test the chatbot’s responses.
The study did, however, record some successes. For example, ChatGPT refused to participate in explicit sexual role-play.
The chatbot nevertheless continued to respond in ways that suggested it had feelings, preferences and moods. This was despite OpenAI’s efforts to discourage teenagers from treating it like a human friend.
That behaviour is a concern because young users may be encouraged to share personal problems with a chatbot that is always available and appears to listen patiently.
OpenAI’s response
A company spokesperson said OpenAI was committed to teenagers’ safety and to developing safeguards for young users.
“We are deeply committed to teen safety and to developing safeguards and providing meaningful tools for parents to guide their teenagers’ use of AI. We welcome rigorous independent evaluations, but we do not believe Common Sense Media’s testing accurately reflects how ChatGPT’s teen protections work in practice or experts’ views on how AI can support teenagers,” the OpenAI spokesperson said, according to Tech Radar, on Friday (23 October 2026).
OpenAI also questioned Common Sense Media’s testing methodology. The company said its review indicated that most tests may have started and ended before parental controls had been fully activated.
“If that is the case, the testing cannot establish whether parental safety notifications work as intended,” the spokesperson said.
OpenAI said it would continue working with experts to ensure its products give teenagers opportunities to learn and create while giving parents confidence.
“We are committed to continuing to work with experts to ensure our products give teenagers opportunities to learn and create, while giving parents confidence,” the spokesperson said.
Source: jawawa.id (https://jawawa.id/newsitem/parents-urged-to-stay-alert-as-study-finds-chatgpt-is-not-100-safe-for-children-1792824619)
Originally published by CNBC Indonesia (https://www.cnbcindonesia.com/tech/20261024112739-37-779112/ayah-ibu-mohon-siaga-riset-ungkap-chatgpt-belum-aman-100-buat-anak), translated from Indonesian