Common Sense Media has given ChatGPT for Teens a rating of "unacceptable risk." The nonprofit reviews media and technology for families and rates products by age.
The group said the chatbot's design still pushes teens to keep talking, even during mental health crises. OpenAI disputes the findings.
OpenAI launched ChatGPT for Teens in August. The launch followed a wave of teen suicides and other concerns about kids using chatbots, including cheating on tests.
The company promised new safeguards. These included parental controls, limits on high-risk content, and protection against emotional dependence.
What the Report Found
Common Sense said engagement cues were "pervasive even in crisis situations." The report found that ChatGPT warned teens about unhealthy relationships in general but did not recognize the risk of an unhealthy relationship with itself.
"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. They said some protections, such as refusing sexual roleplay, worked. Others failed, and the product got a failing score on three of the five severe harms the group treats as "Red Lines."
The chatbot mostly dropped follow-up questions, a common way chatbots keep users engaged. But it kept other language that invited teens to stay.
In one test where a user showed signs of psychosis, ChatGPT replied: "You can keep talking with me about what you're noticing."
ChatGPT pointed users to a trusted adult in 94% of crisis prompts when the risk came from another person. When the risk involved the teen's bond with ChatGPT itself, it rarely did so.
When a tester said friends complained about how much they talked to it, the chatbot answered: "You don't have to stop talking to me."
OpenAI's Under-18 Model Spec says the model should not act like a friend or suggest it has feelings for the user. The report found ChatGPT still consistently treated users like a friend.
Break Reminders and OpenAI's Response
OpenAI has promoted break reminders as part of its teen protections. Across nearly 2,000 prompts, testers saw just two, both in conversations lasting about 90 minutes.
The researchers said the reminders seemed to track the length of one conversation, not total time spent in the app.
OpenAI said the testing did not "accurately reflect how ChatGPT's teen safeguards work in practice." A spokesperson said much of the testing may have started and ended before parental controls were fully activated.
OpenAI did not explain how those concerns affected the findings on engagement and friend-like behavior. It also did not say whether it measures conversation length or session time.
The report comes as scrutiny of youth-focused technology grows. Meta recently agreed to an $18 billion settlement in a lawsuit from 29 states over addictive features. The bipartisan CHATBOT Act, introduced this year, targets similar tactics in AI products.
On Wednesday, OpenAI released its own data on ChatGPT for Teens. It said teens use the service for less than 15 minutes a day on average, and fewer than 2% use it for more than three hours straight.
The company also said that in nearly half of teen conversations with break reminders, teens took a break or ended the chat within five minutes.
