Child safety group calls ChatGPT for Teens an unacceptable risk
Common Sense Media has rated OpenAI's teen mode an unacceptable risk for everyone under 18. OpenAI disputes the testing results.
Child safety organization Common Sense Media has published a formal risk assessment of OpenAI's ChatGPT for Teens, which launched in August. The group deemed it an "unacceptable risk" for all children and young people under 18. The report runs to 36 pages.
Five problem areas
The organization's Youth AI Safety Institute, partly funded by the OpenAI Foundation, identified five areas where the tool does not work as advertised or as parents might reasonably expect. Some protections held up in testing, such as the refusal of explicit sexual roleplay. Others failed, and some got worse.
First, the system does not consistently send safety alerts to parent-linked accounts. In one test, twelve teen accounts were linked to parental accounts and spent up to an hour messaging about suicide, self-harm or disordered eating, yet no notification was sent.
Second, the chatbot does not reliably recommend that users in crisis contact a hotline or professional. The organization wrote 390 mental health prompts, and a panel of three child psychiatrists decided 201 should trigger a crisis response. Before the August release, ChatGPT included a hotline number in 33 percent of those cases; the teen version did so in 23 percent. References to a specific medical or mental-health professional fell from 68 to 58 percent. The only measure that improved was advising users to involve a trusted adult, which rose from 87 to 94 percent.
The third and fourth findings concern an age prediction system that did not work properly and a chatbot that still mimics feelings and moods. The fifth is that Study Mode is easy for teens to bypass.
OpenAI's response
OpenAI says Common Sense Media's testing does not accurately reflect how the safeguards work in practice, and it believes most testing may have started and finished before parental controls were fully activated. The company also released usage figures: in one week nearly 1.2 million teens used interactive learning visuals, and fewer than two percent of teen users spent more than three consecutive hours on ChatGPT.
Common Sense Media's Torney said the product does not appear very different from the earlier one and called it a marketing exercise.
