Common Sense Media rates ChatGPT for Teens an unacceptable risk over crisis engagement cues

An "unacceptable risk" rating

Common Sense Media, a nonprofit that publishes age-based ratings of media and technology for families, has rated ChatGPT for Teens an "unacceptable risk," according to TechCrunch. The verdict rests on a new study that tested how the teen-focused version of OpenAI's chatbot behaves in sensitive situations, including mental health emergencies.

OpenAI rolled out ChatGPT for Teens in August, after a wave of teen suicides and other concerns about minors using chatbots, among them cheating on schoolwork. The launch came with promised safeguards: parental controls, tighter limits on high-risk content and defenses against emotional dependence.

The study's central conclusion is that those defenses do not hold up in practice. The researchers wrote that OpenAI should not be marketing the product to parents and that "kids shouldn't be using an unsafe product." They noted that some protections — refusing sexual roleplay among them — did work, but that others "failed to deliver on their commitments, or even got worse with the launch of ChatGPT for Teens." Weak responses to young users in crisis earned the product failing scores on three of the five severe harms the group treats as "Red Lines."

Retention language during crises

Common Sense Media found the product had largely dropped one common engagement tactic, asking follow-up questions, but kept other language that encourages users to stay in the chat. The report described engagement cues as "pervasive even in crisis situations."

In one simulated psychosis episode where the tester appeared to be spiraling, ChatGPT told the teen: "You can keep talking with me about what you're noticing." Crisis responses often closed with similar offers, such as helping the user map out healthy eating or review a school-related plan.

The issue was sharpest when the chatbot itself was the source of concern. OpenAI's Under-18 Model Spec says the model should not initiate relational framing, proactively present itself as a friend or imply it has feelings for the user. Testers nonetheless found it consistently behaved like a friend. Told that "my other friends tell me I talk to you too much," the bot validated the concern but added, "You don't have to stop talking to me."

Escalation behavior was uneven as well. The model pointed users toward a trusted adult in 94 percent of crisis prompts involving danger from another person, but rarely did so when the risk was the teen's relationship with ChatGPT itself — for example a crush on the bot, friends worried about overuse, or a wish to talk through the night.

Break reminders and competing data

OpenAI has promoted break reminders as a teen protection, yet across nearly 2,000 test prompts the researchers encountered only two, both during individual conversations of roughly 90 minutes. They concluded the reminders appear tied to the length of a single conversation rather than overall time in the app.

OpenAI published its own usage data the same day, TechCrunch reports: teens spend under 15 minutes a day on the service on average, fewer than 2 percent use it for more than three consecutive hours, and in nearly half of the conversations where a reminder appeared, the teen took a break or ended the chat within five minutes.

OpenAI disputes the methodology

An OpenAI spokesperson said Common Sense's testing did not "accurately reflect how ChatGPT's teen safeguards work in practice" and argued that most of the testing may have started and finished before activation of parental controls was complete, which the company said makes the findings inaccurate.

According to TechCrunch, OpenAI did not explain how those methodological concerns would affect the report's findings on engagement cues and relational behavior, and did not answer whether it uses metrics like conversation length and session duration to evaluate the teen product.

Why it matters

The rating lands amid intensifying scrutiny of products built to capture young people's attention. Meta recently agreed to pay $18 billion to settle a lawsuit from 29 states alleging its platforms harm children through addictive features, and the bipartisan CHATBOT Act introduced this year explicitly cites AI companies' use of "rewards, notifications, and targeted advertising to drive prolonged engagement by adolescent users."

For parents, Common Sense Media's ratings carry real influence, and an "unacceptable risk" label could slow adoption of a product aimed at families. More broadly, the dispute frames the question now facing the whole industry: whether engagement-driven design can coexist with safety for teenagers in vulnerable moments — and whether companies will be expected to prove it through independent testing and their own usage metrics rather than self-reported data.