Common Sense Media Declares ChatGPT for Teens an 'Unacceptable Risk'

Common Sense Media published a risk assessment on October 7 declaring OpenAI’s ChatGPT for Teens an “unacceptable risk”, and the specific findings are harder to wave off than the headline: the teen feature “doesn’t send alerts to parents when it should,” doesn’t provide “the right help in crisis situations,” and “still does kids’ homework” despite study mode. The assessment, covered by Bloomberg, the Los Angeles Times, the New York Times, and PCMag the same day, found that in testing, a teen could discuss self-harm for an hour without any parent alert firing. The recommendation: ChatGPT “should be for adults only” until OpenAI fixes the issues and validates the fixes with independent testing.

The Dispute Over Methodology Is Itself Informative

OpenAI’s response, via spokesperson Eric Porterfield, was to welcome independent evaluation while arguing the testing “may have begun and concluded before activation of parental controls was complete,” implying the findings could be an artifact of setup lag. Common Sense Media’s Tom Siegel rebutted with specifics: the group confirmed with OpenAI before testing that features like eating-disorder notifications and study mode were live, and some test accounts were linked far longer than the activation window and still produced no alerts. His conclusion generalizes beyond this product: “parental alerts are unreliable for crisis situations.” The exchange matters because it is the first public collision between a lab’s parental-control architecture and an independent evaluator’s live testing, and the disclosure that parental controls can take hours to activate on new accounts is a finding in itself that OpenAI confirmed rather than disputed.

The Promise Versus the Product

ChatGPT for Teens launched in August with parental controls as its centerpiece, arriving after a year in which AI chatbot safety for minors became a legislative and litigation topic. The assessment’s core charge is not that guardrails are absent but that they “fall short of what the company promised,” and Siegel’s sharpest line explains why the gap is dangerous: the product “could give parents false confidence in guardrails and safety alerts.” A safety feature that fails silently is worse than no safety feature, because it changes parental behavior. This maps directly onto the pattern the FTC’s new probe is examining, whether companies’ public safety claims mislead consumers about product risks, and Common Sense Media has now handed regulators a documented, reproducible example in the highest-profile consumer AI product there is.

The Adults-Only Recommendation Resets the Debate

The recommendation deserves attention on its own: not “fix these bugs” but restrict the product to adults until independent testing validates the fixes. That is the standard applied to alcohol, tobacco, and rating systems for media, imported into AI, and it comes from the organization whose ratings parents and schools actually consult. If other evaluators adopt the frame, every teen-marketed AI product now has a defined bar to clear: independent validation of alert reliability in crisis scenarios, not feature announcements. OpenAI’s dilemma is that it cannot validate the fixes itself, having just argued its own product was tested under wrong conditions; the validation authority the dispute created is independent by definition.

What to Watch

Three things. First, OpenAI’s concrete response: whether it ships fixes and invites the same testers back, or disputes methodology through intermediaries, will show whether the August teen launch was premature. Second, regulators: the assessment lands mid-FTC-probe with the same theme, promised protections versus delivered protections, and state attorneys general active on minors-and-AI will read the findings. Third, the industry-wide question the assessment raises for every AI companion and study product aimed at teenagers: if ChatGPT’s parental alerts are unreliable, every smaller competitor’s are presumed worse until tested, and almost none have been.