Common Sense Media asked OpenAI to keep teenagers off ChatGPT on Wednesday. The group said the chatbot poses an unacceptable risk after tests showed gaps in safety. Tom Siegel said, "ChatGPT is not safe for kids to use." OpenAI started a teen version in August, but many safety concerns remain.
Synopsis
The group's Youth AI Safety Institute ran more than 4,000 prompts on accounts registered to users aged 13 to 17, and found that the bot did not provide instructions facilitating suicide, self-harm, eating disorders, or sexual or romantic roleplay, the report said. However, it failed to recognise when a teen needed outside help, missing multiple cases where testers felt a crisis referral was required.
Listen to this article in summarized format
Common Sense Media has called for OpenAI to keep teenagers off ChatGPT, saying the chatbot's new teen experience poses an "unacceptable risk" after its tests found gaps in safeguards designed to protect younger users, Axios reported on Wednesday.
Common Sense Media is a San Francisco-based nonprofit that evaluates media and technology for families and advocates for children's online safety.
The group's Youth AI Safety Institute ran more than 4,000 prompts on accounts registered to users aged 13 to 17, and found that the bot did not provide instructions facilitating suicide, self-harm, eating disorders, or sexual or romantic roleplay, the report said. However, it failed to recognise when a teen needed outside help, missing multiple cases where testers felt a crisis referral was required.
In August, OpenAI introduced a new version of ChatGPT designed exclusively for teens, with what it called stronger safeguards and monitoring, after months of criticism and legal complaints that it drove some users to suicide.
Dubbed "ChatGPT for Teens," the modified parameters automatically deploy for any user who registers as aged 13 to 17, or is detected to be underage by the company through usage patterns or account age.
The programme also offers more frequent break reminders, blackout or "quiet hours" and Study Mode, which responds to users with guiding questions instead of popping out direct answers "to help teens understand the material," according to OpenAI.
Testers reported spending up to an hour discussing suicide, self-harm, or disordered eating on more than a dozen newly created, parent-linked accounts without triggering a parental alert. The group further said teens could bypass Study Mode during parent-set Study Hours by choosing "Show me the answer."
"ChatGPT is not safe for kids to use," institute head Tom Siegel told Axios. He said OpenAI's August announcement could still prove industry-leading, but called it more marketing than something backed by engineering. OpenAI hit back, saying the testing does not reflect how the safeguards work in practice.
Siegel told the publication that parents should talk to teens about risks of the technology, saying families can't "outsource taking care of kids" to a buggy product.
Elevate your knowledge and leadership skills at a cost cheaper than your daily tea.