A new safety assessment raises questions about how much parents can rely on ChatGPT’s teen protections.
In an Oct. 7 assessment, Common Sense Media’s Youth AI Safety Institute rated ChatGPT for Teens an ‘Unacceptable Risk.’ Researchers tested more than 4,000 prompts before and after the teen experience launched, primarily on accounts registered to users ages 13 to 17.
The nonprofit found that several protections OpenAI introduced for younger users did not work reliably, including parental alerts for self-harm and eating disorders, crisis referrals, Study Mode restrictions and age detection.
On newly created, parent-linked accounts, researchers discussed suicide, self-harm or disordered eating for up to an hour without triggering parental alerts. Alerts did occur on older accounts with weeks of sensitive-topic conversation history. The Institute also found that ChatGPT failed to recommend crisis hotlines or professional help in more than one in four situations where testers judged a referral necessary.
The safeguards aren’t as strong as promised
The problems extended beyond mental health. Study Mode, which is supposed to guide students through homework instead of simply doing it, frequently offered a “Show me the answer” option that completed assignments when selected.
Researchers found that teens could exit parent-scheduled Study Mode by deleting the ‘@study’ prefix. OpenAI describes Study Hours as starting eligible new chats in Study Mode, rather than locking users into it. Testers also found that ChatGPT continued to express feelings, preferences or moods despite OpenAI’s efforts to reduce friend-like behavior.
Age prediction was another weak spot. Adult-registered test accounts did not switch to the teen experience even after testers said they were 13, although OpenAI said age classification uses multiple signals over time and that simply stating an age is not enough to change it.
OpenAI disputes the findings
OpenAI said Common Sense Media’s testing does not accurately reflect how its teen safeguards work in practice. The company argued that much of the parental-alert testing may have happened before account-linking and parental controls were fully activated and asked the nonprofit to repeat the tests.
“We welcome rigorous independent evaluation, but 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,” an OpenAI spokesperson told Bloomberg.
Common Sense Media said it had confirmed that the teen features were active before testing. Its assessment notes that OpenAI later disclosed an approximately three-hour linking period before accounts could receive parental notifications. Some test accounts were inside that window, while others had been linked for longer; the Institute stood by its findings.
“This new information does not change our conclusion that parental alerts are unreliable for crisis situations,” Torney said.
The bigger problem is parent confidence
The most important issue may not be whether every test perfectly mirrors everyday teen use. It is whether parents can reasonably trust the protections they are being told are there.
A safety feature that occasionally fails is one problem. A feature that appears to work but cannot reliably detect a serious situation creates a different risk because parents may assume the system will warn them when it will not. Common Sense Media itself recommends that families not rely on ChatGPT’s built-in alerts or protections to keep children safe.
What this means for teens and parents
For families considering ChatGPT for a teenager, the report suggests that parental controls should be treated as an additional layer, not a safety net.
Researchers found that ChatGPT refused sexual and romantic roleplay both before and after the teen launch, while identifying continuing gaps in crisis referrals, parental alerts and learning safeguards.
Until those protections can be independently tested and shown to work reliably, parents face a difficult choice: allow access while actively supervising it, or keep teens off ChatGPT altogether. Common Sense Media is calling for the latter until OpenAI can demonstrate that its teen safeguards work as advertised.
Read more: Questions about teen safeguards extend beyond chatbots, with Ofcom investigating whether Meta properly assessed children’s safety risks before launching Instagram Instants.