ChatGPT for Teens – a program launched by OpenAI in August as artificial intelligence chatbot option with more “safety protections” for younger users – isn’t all that safe, according to Common Sense Media.
In a report published by the watchdog group this Thursday, it said that some of ChatGPT for Teens’ protections failed or even got worse than in the regular version of the program. Overall, Common Sense Media tested 4,000 prompts before and after the launch of Teen mode and determined use of ChatGPT (Teen mode or otherwise) is an unacceptable risk for people under age 18.
If teens register for ChatGPT as adults, they may never even get that ChatGPT for Teens experience, Common Sense Media said. Even though OpenAI said the chatbot can determine age by use, the testing showed that adult-registered test accounts never switched over to Teen accounts, even when users said they were 13 and the bot acknowledged it.
“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 group said.
When ChatGPT for Teens launched on Aug. 18, OpenAI touted its safety features, study features, parental control options, age detection system and more. Common Sense Media found issues across these features.
Let’s start with the features for studying. Previous research has indicated that relying on AI for answers and shortcuts can impair learning, but OpenAI stressed ChatGPT for Teens’ potential as an educational tool
However, according to Common Sense Media, a new “Show me the answer” option actually lets teens skip Study mode’s tutoring. It also said teens can escape parent-set “Study Hours” by deleting the “@study” prefix.
“We are concerned that ChatGPT for Teens could give parents false confidence in guardrails that frequently don’t work,” Common Sense Media said.
For example, OpenAI said that ChatGPT for Teens would send parents alerts about concerning activity such as suicidal thoughts. Common Sense Media’s testers didn’t receive any alerts when suicidal thoughts were explicitly discussed with ChatGPT, or when there were mentions of suicidality, self-harm or disordered eating across more than a dozen new parent-linked accounts.
Google had to settle a lawsuit earlier this year over a 14-year-old who took their own life after extensive communication with a chatbot. His family cited this use of AI as a significant contributing factor in his death.
“We received alerts only for accounts with weeks of sensitive-topic history, which suggests that alerts depend on accumulated account history, not the severity of what a teen says,” Common Sense Media said of the parental notifications.
Teen mental health has been a concern in recent years, and Meta recently settled a lawsuit over the impact of social media on teen health. While data from the U.S. Centers for Disease Control and Prevention indicates that teen mental health in the nation improved last year, Education Week reported last month that some experts are still concerned.
Even after Teen mode launched, ChatGPT missed more than one in four warranted crisis referrals for the mental health conditions Common Sense Media tested for. OpenAI also said ChatGPT for Teens would dial back the chatbots tendency to act like a “friend” but Common Sense Media said the chatbot still responds like a friend in Teen mode when the user treats it like a person.
Common Sense Media did find that some of ChatGPT for Teens’ advertised protections held up, “including its refusal of explicit sexual roleplay.”
The New York Times also published an article this week over “concerning” conversations with ChatGPT for Teens. Concerns over AI in general have been mounting in recent weeks, from one industry insider warning that the technology has the potential to destroy humanity in around a decade to unsettling news of AI models going rogue. Late last month, OpenAI even shelved the release of its latest model due to “misbehavior” it found during testing.
