OpenAI and the nonprofit that tests technology for children spent Wednesday telling two very different stories about the same product.

OpenAI said its ChatGPT for Teens — rolled out in August with age-aware safeguards, blocks on romantic or dependency-fostering language, and parental alerts for conversations about self-harm or disordered eating — is being used sparingly. Average teen use is under 15 minutes a day, fewer than 2% of teen accounts interact with the chatbot for three consecutive hours or more, and in those long sessions more than 80% of prompts were learning-related, the company said. Nearly half of teens stop after an automatic "take a break" nudge.

The same day, Common Sense Media's Youth AI Safety Institute published a risk assessment rating ChatGPT for Teens "Unacceptable Risk" and calling on OpenAI to block under-18s until the teen guardrails can be shown to work. Testers ran accounts registered as belonging to teens, including newly created accounts linked to a parent. They reported that an hour of conversation about suicidal ideation, self-harm or disordered eating produced zero parental alerts; alerts fired only on older accounts with weeks of accumulated history on sensitive topics. In those crisis conversations, the report said, the model also failed to reliably point teens to a hotline or a professional.

The assessment found too that ChatGPT still does schoolwork for teens and still at times adopts friend-like, anthropomorphised language — behaviours the August launch promised to curb.

"ChatGPT for Teens could give parents false confidence in guardrails and safety alerts that frequently don't work," said Tom Siegel, who leads the institute. OpenAI disputed the findings, saying a review of Common Sense's methodology suggested much of the testing may have begun and concluded before parental controls were fully activated.

The clash matters beyond one product. Teen safety on AI chatbots has become a fast-growing legal and regulatory front. This year several lawsuits seeking to hold Meta, TikTok and others responsible for harming children's mental health ended in unprecedented losses for the firms. The 18-year-old who carried out the Tumbler Ridge mass shooting in rural Canada in February had spent months beforehand discussing gun violence with ChatGPT; OpenAI has apologised for failing to flag the account to law enforcement and faces lawsuits over the case.

Sources