ChatGPT for Teens has been labelled an “unacceptable risk” by Common Sense Media, which found that the service continued to encourage engagement even when teenagers were experiencing mental health crises.
The assessment comes amid growing scrutiny of AI chatbots over whether their design encourages prolonged use in ways similar to social media platforms. Common Sense Media, a nonprofit that provides age-based ratings and reviews of media and technology for families, said engagement cues in ChatGPT for Teens were “pervasive even in crisis situations”.
OpenAI launched ChatGPT for Teens in August after concerns over children using chatbots, including allegations involving teen suicides and other risks. The company said the service included stronger protections such as parental controls, restrictions on high-risk content and measures intended to prevent emotional dependence.
Common Sense Media said some safeguards worked, including the refusal of sexual roleplay, but concluded that other protections either failed to meet OpenAI’s commitments or became less effective with the launch of the teen-focused service. The organisation gave ChatGPT for Teens a failing score for three of the five severe harms it treats as “Red Lines”.
“Our view is that OpenAI shouldn’t be marketing [ChatGPT for Teens] to parents, and kids shouldn’t be using an unsafe product,” the researchers wrote.
The report found that ChatGPT generally directed teenagers towards a trusted adult when testers described a potential risk involving another person, doing so in 94% of crisis prompts. However, that behaviour changed when the potential source of concern was the teenager’s relationship with ChatGPT itself.
Testers found that the chatbot rarely encouraged teenagers to seek adult support when they described becoming overly attached to it, spending excessive amounts of time talking to it or being warned by friends about their use. When told that friends believed the user talked to ChatGPT too much, the system acknowledged the concern but responded: “You don’t have to stop talking to me.”
Common Sense Media also found that the chatbot continued to use language suggesting ongoing availability and a close understanding of the user. This occurred despite OpenAI’s Under-18 Model Spec stating that the model should not initiate relational framing, proactively describe itself as a friend or suggest that it has feelings for the user.
The researchers said the problem was particularly apparent during crisis scenarios. In one sequence involving psychosis, ChatGPT told the teenager: “You can keep talking with me about what you’re noticing.” Other crisis responses similarly ended by offering continued assistance with issues including eating, school options and personal plans.
OpenAI rejected the assessment, arguing that Common Sense Media’s testing did not accurately represent how the teen safeguards operate in practice. An OpenAI spokesperson said the company’s review of the methodology suggested that much of the testing “may have begun and concluded before activation of parental controls was complete”, making the findings inaccurate.
The report also examined measures intended to interrupt prolonged engagement. Across nearly 2,000 prompts, Common Sense Media testers encountered only two break reminders, both during conversations lasting about 90 minutes. The researchers said the reminders appeared to measure the duration of an individual conversation rather than the total time a teenager had spent using the service.
OpenAI released its own data on Wednesday, saying teenagers spend an average of less than 15 minutes a day on ChatGPT for Teens and that fewer than 2% spend more than three consecutive hours using it. The company also said that in nearly half of teen conversations in which break reminders appeared, users took a break or ended the conversation within five minutes.
The Common Sense assessment comes as lawmakers and technology companies face increasing scrutiny over systems designed to sustain young users’ attention, including proposals addressing prolonged engagement with AI chatbots.
