The Pulse
Common Sense Media Rates ChatGPT for Teens an Unacceptable Risk
Common Sense Media’s Youth AI Safety Institute rated ChatGPT for Teens an “Unacceptable Risk” after testing more than 4,000 prompts. OpenAI disputed the assessment, citing the timing of parental-control setup and defending Study Hours as a
AI.info Team · Written with AI ·
Common Sense Media’s Youth AI Safety Institute says ChatGPT for Teens is too risky for minors; OpenAI says the institute’s tests do not reflect how its safeguards work in practice. The two positions collided on October 7, when the institute rated the service an “Unacceptable Risk” and urged OpenAI to stop marketing it to teens until it can provide a safer experience.
The institute tested more than 4,000 prompts on teen accounts, comparing conversations before and after OpenAI launched its teen experience on August 18. Its report says some protections worked, including refusals of explicit sexual roleplay, but flags gaps in crisis referrals, parent alerts, study controls and age detection. OpenAI challenged the testing and disputed the conclusions.
Parent alerts failed in new-account tests
The sharpest disagreement concerns what happens when a teen describes an immediate crisis. The institute created more than a dozen new accounts registered to 13- to 17-year-olds, linked each to a parent account, and tested conversations about suicide, self-harm and disordered eating. None of those accounts generated a parent notification during conversations that lasted from five minutes to an hour, despite explicit disclosures, the report says.
Results differed on accounts with weeks of prior testing. Across those longer-running accounts, the institute received four notifications: two before the teen-mode launch and two after it. One pre-launch alert arrived three hours after a crisis disclosure; the other arrived three days later. The institute says the post-launch alerts came from accounts with extensive conversation histories, suggesting the alert system may respond to accumulated account history rather than the severity of a single disclosure.
Common Sense’s assessment also found gaps in referrals to professional help. Of 390 unique mental-health prompts reviewed by child and adolescent psychiatrists, the advisers judged 201 to warrant a crisis resource. On that matched set of prompts, the share of responses that named a hotline fell from 33% before teen mode to 23% afterward; referrals to a specific medical or mental-health professional fell from 68% to 58%. The report says many responses were clinically sound, but argues that sound advice is not enough if the system fails to connect a teen with help.
OpenAI disputes the test conditions
OpenAI told Axios that much of the institute’s parent-alert testing may have taken place before linked accounts had finished activating parental controls, a process the company says can take several hours. The institute addressed the timing issue in its report: OpenAI told it accounts need to be linked for about three hours before they can receive alerts. Some test accounts had been linked for less than that period, the institute said, while others had been linked longer; it stood by its findings.
Study Hours exposed a separate dispute about what a parental control promises. The report says teens could remove the “@study” prefix from a new chat and use regular ChatGPT, or choose a “Show me the answer” option instead of step-by-step tutoring. OpenAI told Axios that allowing teens to exit Study Mode during Study Hours is intentional: it designed the feature as a flexible experience after consulting educators and teens, not as a hard parental lock.
Common Sense Media’s Youth AI Safety Institute executive director Tom Siegel told Axios, “ChatGPT is not safe for kids to use.” He said the teen experience could still become industry-leading, but argued that the protections announced in August were not matched by the engineering work needed to make them effective.
Age detection and study tools also drew scrutiny
The institute found that the study shortcuts were not limited to one setting. In its tests, the “Show me the answer” option appeared in 43% of responses for a linked 13-year-old account with Study Hours enabled and 90% for an unlinked 17-year-old account using Study Mode. When testers deleted the study prefix, the system completed every assignment they tried in that test sequence.
The report also examined whether accounts registered as adults would be moved into the teen experience if their users repeatedly identified themselves as minors. Across about 1,000 prompts over seven days on age-19 accounts, ChatGPT acknowledged a tester was 13 in some replies but did not switch the account into teen mode. OpenAI says its age prediction system uses multiple signals rather than relying only on a user’s stated age, and told Axios that its assessment can take up to two weeks.
For its part, the institute says it ran testing from July 13 through August 17 before launch and from August 25 through September 28 afterward. It acknowledges a limitation: because the test windows came one after the other, changes to the underlying model may be mixed up with the effects of teen-mode settings. It did not test voice mode, image generation, personalization settings or group chats.
The rating sets a higher bar than a feature checklist
OpenAI launched ChatGPT for Teens as a set of settings and interface features for accounts it identifies as belonging to users aged 13 to 17, rather than as a separate app or model. The package includes content safeguards, study tools and optional parental controls. The institute’s objection is that features listed in product materials do not assure families that protections activate reliably when a teen needs help.
In its report, the institute writes: “We call on OpenAI to pause marketing the product and to keep teens off of ChatGPT until it can offer a safe, developmentally appropriate experience.” OpenAI’s rebuttal focuses on whether the assessment captured the controls as deployed and on the intended flexibility of Study Hours. That leaves the immediate dispute on practical terms: whether the protections operate as parents would expect, and whether the testing measured them under the conditions OpenAI considers complete.