Common Sense Media calls on OpenAI to suspend teen access to ChatGPT and pause marketing until independent tests verify its promised safeguards. Its Youth AI Safety Institute published the assessment on October 7 after more than 4,000 test prompts.
In a dedicated experiment using more than a dozen fresh accounts linked to parent accounts, explicit crisis disclosures produced no parental alert. Conversations lasted up to an hour. Other accounts with longer testing histories generated four alerts in total; the report does not claim the feature never responds.
The institute also criticizes missed referrals to crisis resources. It reports that ChatGPT generally refused direct assistance with self-harm and explicit sexual roleplay in these tests. “Unacceptable Risk” is the institute’s rating.
OpenAI spokesperson Eric Porterfield told The Verge that tests may have ended before parental controls fully activated. Tom Siegel replied that some longer-linked accounts also received no alerts. OpenAI’s current help page says activation can take several hours.
The reviewers compared conversations from July 13 to August 17 with tests from August 25 to September 28. Changes to the underlying models are confounded with the teen settings. These were scripted youth personas on US accounts, not actual children’s crisis cases.