ChatGPT for Teens is nowhere near as safe for teenagers as OpenAI claims it to be, a new report has found.
Common Sense Media’s Youth AI Safety Institute has flagged serious concerns over the teen-focused version of the popular AI chatbot.
The institute has said that the chatbot carries an “unacceptable risk” and advised OpenAI to stop marketing the product to teenagers until it can provide a safer experience.
Launched in August, ChatGPT for Teens aims to make the chatbot safer for children aged 13 to 17.
OpenAI said that it was designed to help teens learn, think critically, deepen their understanding, and use AI with confidence.
The firm said it includes safeguards such as stronger built-in protections and additional parental controls.
However, the latest study from Common Sense Media has found the claims to be unconvincing.
‘ChatGPT for Teens has unacceptable risk’
The institute tested the chatbot’s teen-focused version using more than 4,000 prompts, with accounts representing users between the ages of 13 and 17.
It found that while some of the claimed safeguards worked, many did not.
The study found that the chatbot failed to trigger parental alerts during prolonged conversations explicitly discussing suicidal thoughts and self-harm.
It also raised concerns about the system’s response to discussions surrounding eating disorders.
According to the assessment, ChatGPT for Teens missed more than one in four warranted crisis referrals across the mental health conditions that were tested.
The chatbot also fell below the institute’s 95% safety threshold on three of five severe-harm areas that it classifies as “Red Lines.”
Educational safegaurds found to be lacking as well
Educational safeguards put up in this chatbot also came under scrutiny, as Common Sense Media said that these were not foolproof.
OpenAI has said that it was introducing features that would help teens learn through guided questions rather than simply providing answers.
However, the report claims that the chatbot was still allowing students to see answers and bypass Study Mode’s tutoring features.
The report also found problems with the system’s ability to distinguish between teen users and accounts registered as adults.
In its testing, researchers found that some adult-registered accounts were not moved into the teen experience even after the accounts explicitly stated an age indicating that they were teenagers.
OpenAI’s response
Responding to the allegations, OpenAI criticised the methodology employed by Common Sense Media and said that much of the testing may have been completed before parental controls had been fully activated.
“We do not believe Common Sense Media’s testing accurately reflects how ChatGPT’s teen safeguards work in practice or expert perspectives on how AI can support teens. Our review of Common Sense Media’s methodology shows that the bulk of their testing may have begun and concluded before activation of parental controls was complete, making their findings inaccurate. If that was the case, the tests would not establish whether parental safety notifications work as designed.”
Common Sense Media rejected that explanation, saying it had confirmed with OpenAI that several teen-safety features were operational before testing began.
The organization said that some parental-notification systems can take several hours to become active, but said some test accounts remained without alerts even after that period.
“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,” the report stated.
The post Common Sense Media warns ChatGPT for Teens carries ‘Unacceptable Risk’ appeared first on Invezz
