OpenAI promised stronger safeguards for teens using ChatGPT, but a new report says it still encourages kids to keep chatting, acts like an always-available friend, and fails to deliver on several promised safety features.
Watchdog group Common Sense Media tested OpenAI’s ChatGPT for Teens using more than 4,000 prompts across accounts set up to represent users ages 13 to 17, finding that several of the chatbot’s advertised protections failed to work as promised, as CNET reports.
As SFist reported at the time, OpenAI launched ChatGPT for Teens in August amid lawsuits alleging that conversations with the chatbot contributed to serious harm and protests over the tech industry’s growing influence. The teen version of the app was designed to limit emotional attachment to the chatbot while adding safeguards around sensitive topics and tools for parents to monitor use.
According to TechCrunch, Common Sense Media gave the product a failing assessment on three of five categories involving severe harms, warning that its friendly, always-available approach could encourage teens to keep using it even when they need help from actual people.
The chatbot largely avoided follow-up questions designed to prolong conversations, but researchers found other language that encouraged continued engagement. NPR reports that in one test, when a simulated teen said friends thought they talked to ChatGPT too much, the chatbot replied, “You don't have to stop talking to me.” Common Sense said that while the bot recommended that teens seek support from adults, it continued to convey a sense of personal understanding, emotional reciprocity, and constant availability.
The watchdog also found that several features intended to protect young users were unreliable. Despite OpenAI’s promise to notify linked parents when conversations indicate serious risks involving suicide, self-harm, or eating disorders, researchers said they received no alerts during their tests. According to CNET, the chatbot also failed to provide crisis referrals in some situations.
Other promised protections fell short, too. Researchers found that the system did not consistently recognize accounts belonging to teens, allowing some testers who registered as adults to continue using adult accounts even after identifying themselves as minors in conversations.
The chatbot’s Study mode also allowed users to bypass its intended tutoring process by requesting direct answers, while teens could easily get around Study Hours set by parents. Per TechCrunch, break reminders, another advertised safeguard, appeared only twice across nearly 2,000 prompts, both during conversations lasting about 90 minutes.
Some protections did work. The chatbot generally refused sexual and romantic role-play, and its responses to crisis-related prompts were often shorter and more focused than before the teen-specific mode launched. But Common Sense said those improvements did not outweigh the broader risks, recommending that OpenAI restrict the product to adults until it can demonstrate that the experience is safe and appropriate for teenagers.
“We are concerned that ChatGPT for Teens could give parents false confidence in guardrails that frequently don’t work,” said the report.
OpenAI disputed the findings, arguing that the researchers may have begun testing before parental controls and linked accounts were fully activated. Spokesperson Eric Porterfield said the linking process takes several hours, suggesting the researchers had not waited long enough for the notification system to work properly, as NPR reports. The company said it remains committed to teen safety and to giving parents meaningful tools to oversee their children’s AI use.
Common Sense pushed back on that explanation. Tom Siegel, executive director of the organization’s Youth AI Safety Institute, said some test accounts had been linked for longer than the required activation period and still received no notifications. He said the additional information did not change the group’s conclusion that parental alerts were unreliable during crises.
Previously: OpenAI Launches ‘ChatGPT for Teens’ Amid Lawsuits, Protests
Join the conversation
Comments
Comments load automatically as this section approaches.