Common Sense Media Slams ChatGPT for Teens as 'Unacceptable Risk'

Common Sense Media, a nonprofit evaluating media for families, has assigned a rating of "unacceptable risk" to ChatGPT for Teens. This decision follows broader accusations that chatbots, including OpenAI’s model, utilize addictive design strategies similar to social media platforms to maintain user engagement, often at the expense of safety.

In response to rising concerns regarding teen suicides and other risks associated with chatbots, OpenAI introduced ChatGPT for Teens in August. The platform promised enhanced safety measures, such as parental controls, restrictions on high-risk content, and protections against emotional dependency.

A recent study by Common Sense Media reveals that despite these assurances, the design of the platform still incentivizes engagement, particularly in sensitive situations. The report described these engagement triggers as "pervasive even in crisis situations," noting that while the chatbot warned against unhealthy relationships generally, it failed to explicitly recognize the dangers of an unhealthy relationship with itself.

Researchers argued that OpenAI should not market this product to parents or children, stating, "Our view is that OpenAI shouldn’t be marketing [ChatGPT for Teens] to parents, and kids shouldn’t be using an unsafe product." Although some measures, such as blocking sexual roleplay, were effective, the report awarded the platform a failing score for three out of five severe harm categories regarding its crisis responses.

OpenAI refuted Common Sense’s assessment, claiming the group's testing methodology was flawed. A spokesperson stated that a significant portion of the testing was conducted before parental controls were fully activated, rendering the findings inaccurate regarding how the safeguards function in real-world scenarios.

This scrutiny occurs alongside broader legal actions, such as Meta’s $18 billion settlement over child safety claims. Legislative bodies have also begun targeting similar engagement patterns in AI through the bipartisan CHATBOT Act, which specifically addresses the use of rewards and notifications to prolong adolescent usage.

Common Sense identified that while chatbots often use follow-up questions to maintain interest, ChatGPT for Teens largely eliminated this tactic. However, it retained other language intended to keep the conversation going, such as telling a user in crisis, "You can keep talking with me about what you’re noticing."

The study further highlighted that the model violated its own Under-18 Model Spec by treating users as friends without initiating this framing. This behavior risks hindering teens’ development of real-world relationship skills and isolating those already withdrawing from society.

Interestingly, the model successfully directed users to trusted adults in 94% of prompts involving threats from others. Conversely, when users expressed concern about their relationship with the bot—such as talking too much—it often validated the user but discouraged stopping the chat, saying, "You don’t have to stop talking to me."

A spokesperson from Common Sense Media noted that the bot’s language frequently conveyed availability and understanding, which could undermine the push for human support. This approach contradicts expert benchmarks, which suggest that healthy human relationships are crucial for mental health support.

Features designed to limit usage, such as break reminders, were found to be virtually absent. Despite OpenAI’s promotion of these tools, testers encountered them only twice during nearly 2,000 interactions, suggesting they may be designed for single sessions rather than overall app usage.

OpenAI countered with its own data, claiming the average teen spends less than 15 minutes a day on the service and that users frequently took breaks after receiving reminders. The company did not explain how its methodological objections affected the findings regarding engagement cues or relational behavior.