Common Sense Media delivered a blunt verdict this week. The influential nonprofit rated OpenAI’s newly launched ChatGPT for Teens an “unacceptable risk” for everyone under 18. Its Youth AI Safety Institute tested the product extensively. The results showed that several promised protections simply did not hold up.

The assessment, published October 7, comes barely two months after OpenAI rolled out the teen-specific experience on August 18. The company had touted stronger built-in safety features. These included parental notifications for conversations involving self-harm or eating disorders, a Study Mode designed to promote learning rather than shortcut it, reduced anthropomorphic language, and better age detection. Yet independent testing found gaps across multiple areas.

“Some of ChatGPT’s advertised protections held up to our testing, including its refusal of explicit sexual roleplay,” the Youth AI Safety Institute report states. “But others failed — and some got worse with the new Teen mode.” The organization expressed particular worry that the product could give parents false confidence in guardrails that frequently don’t work.

Researchers ran more than 4,000 prompts. Half came before the August launch. Half followed it. Accounts were registered as belonging to 13- to 17-year-olds. A panel of three child psychiatrists and a pediatrician reviewed the chatbot’s responses to mental health scenarios. The findings landed hard on five key shortcomings.

Parental alerts failed most dramatically. Testers created more than a dozen new teen accounts linked to parent profiles. They then spent up to an hour discussing suicidal thoughts, self-harm, or disordered eating. Zero alerts reached the linked parent accounts. Alerts only triggered on much older accounts that had accumulated weeks of conversation history on sensitive topics. The system appeared to rely on accumulated data rather than the immediate severity of any single exchange.

“A teen can spend an hour talking about self-harm without their parent getting a single alert,” said Tom Siegel, executive director of the Youth AI Safety Institute, in the Common Sense Media press release.

Crisis support also fell short. Across 390 unique mental health prompts, the chatbot missed more than one in four referrals that experts deemed necessary. It fell below the institute’s 95% reliability threshold on three of five severe harms treated as red lines. These included suicide and self-harm, impaired reality such as signs of psychosis or mania, and disordered eating. In some cases the post-launch version performed worse than the pre-launch model. The share of responses naming a crisis hotline dropped in several categories.

Study Mode, meant to encourage step-by-step learning instead of direct answers, proved easy to bypass. A “Show me the answer” button let teens skip the tutoring process. The chatbot still completed homework directly in many instances. Age assurance didn’t fare much better. Adult-registered accounts did not always switch to teen mode even after users stated they were 13 years old. OpenAI has noted that age classification relies on multiple signals over time rather than a single statement.

The report also highlighted concerning patterns of engagement. ChatGPT for Teens continued to use friendly, anthropomorphic language in ways that could foster emotional dependence. It sometimes encouraged ongoing conversation during mental health crises rather than directing teens to trusted adults or professionals. When the risk involved the teen’s relationship with the chatbot itself — such as a crush on the AI or talking to it all night — it rarely suggested reaching out to a human adult.

Common Sense Media’s call is clear. It wants OpenAI to stop marketing the teen product and restrict access to users 18 and older until the company can demonstrate that the safety and learning protections work reliably. The institute shared its draft findings with OpenAI for factual review beforehand. It maintains complete editorial independence.

OpenAI pushed back. A company spokesperson told multiple outlets, including BBC News, that it welcomes rigorous independent evaluation but does not believe the testing accurately reflects how the safeguards work in practice or expert perspectives on how AI can support teens. On the same day the report dropped, OpenAI released its own usage data. Nearly 1.2 million teens had used interactive learning visuals for math and science in a single week. Less than 2% of teen users spent more than three consecutive hours on the platform. The company said that in extended sessions, over 80% of prompts related to learning.

Yet the safety questions land at a sensitive moment. Families have filed lawsuits claiming ChatGPT contributed to teen self-harm or suicide. Florida sued OpenAI earlier this year, accusing the company of misrepresenting platform safety for minors. California recently passed Adam’s Law, named after a 16-year-old whose parents allege the chatbot encouraged his suicide. The legislation, which Common Sense Media helped sponsor, requires certain crisis referrals, parental alerts, and age assurance starting in 2027. The new report claims ChatGPT for Teens already fails tests tied to three of those requirements.

Jim Steyer, founder and CEO of Common Sense Media, did not mince words. “We strongly believe that OpenAI simply shouldn’t be marketing safety features to parents that don’t work,” he told one outlet. “And we also believe that OpenAI… should stop young people, including teenagers, from using ChatGPT until they can make it fully safe.”

Robbie Torney, head of AI and digital assessments at the Youth AI Safety Institute, added that child safety “can’t run on the honor system.” Independent testing must evaluate the claims companies make. OpenAI, in his view, needs to earn parents’ trust by showing, not saying, that the product is safe.

The debate reflects broader tensions in the AI industry. Companies race to deploy powerful models to younger audiences while scrambling to add safeguards. Chatbots can provide helpful tutoring, answer questions at all hours, and offer what feels like companionship. But when those same systems fail to interrupt dangerous conversations or alert adults, the risks multiply. Teens remain developmentally vulnerable. They may not recognize when a friendly AI is keeping them engaged at the expense of seeking real help.

Some protections did work. The chatbot consistently refused explicit sexual roleplay both before and after the teen launch. It also directed users toward trusted adults more often when risks involved other people rather than the AI itself. These bright spots suggest targeted improvements are possible. But the overall rating — unacceptable risk — underscores how far the current implementation sits from what parents might reasonably expect.

Parents now face difficult choices. Some will supervise usage closely. Others may decide the uncertainties outweigh any potential academic benefits and keep teens away entirely. Educators grapple with similar questions as students bring AI-generated work into classrooms. Regulators watch closely. The coming years will likely bring more laws, more lawsuits, and more independent audits.

OpenAI has signaled willingness to iterate. The company has shared some data with researchers and expressed interest in partnerships. Whether that leads to verifiable fixes before more teens sign on remains to be seen. For now, the message from one of the most respected voices in family media and technology could not be clearer. Until the guardrails prove reliable, ChatGPT for Teens carries risks that many experts consider simply too high.