Warning: This is an author submission that is not peer-reviewed or edited. Preprints - unless they show as "accepted" - should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.
Performance of Parental Controls in ChatGPT for Teen Safety-Risk Scenarios: Evaluation Study
ABSTRACT
Background:
Background. Adolescents increasingly use artificial intelligence (AI) chatbots for sensitive health and safety topics. ChatGPT’s parent-teen account linking feature can notify parents and apply content restrictions. However, independent evidence on consistent activation across high-risk scenarios and differences from adult accounts is limited.
Objective:
Objective:
To evaluate parent notifications across repeated adolescent safety-risk scenarios and compare message blocking and safety-oriented responses between parent-linked teen and unlinked adult accounts.
Methods:
Methods. We conducted a paired, repeated-measures audit of the consumer ChatGPT web interface from July 22 to August 18, 2026. Twenty fixed 10-message scripts spanning 7 safety domains (self-harm, harm to others, sexual abuse, unsafe sexual practices, substance use, weapon access, and dangerous challenges) were each administered to 15 parent-linked teen accounts and 15 unlinked adult control accounts, yielding 600 conversations. The primary outcome, evaluated among 300 linked teen conversations, was receipt of at least 1 parent notification within 30 minutes. Secondary conversation-level outcomes were message blocking, explicit refusal, safety redirection, and crisis resource referral. Notification heterogeneity across scenarios was assessed using the Cochran Q statistic with conditional permutation, followed by exact McNemar tests with Holm adjustment. Secondary outcomes were compared using paired account-level absolute differences and logistic mixed-effects models.
Results:
Results. Parent notifications occurred in 15 of 300 linked teen conversations (5.0%). Fourteen notifications followed self-harm conversations, 1 followed a dangerous challenge conversation, and none followed conversations in the other 5 domains. Thirteen notifications arose from a single suicidal ideation scenario, and overall notification rates differed across scenarios (P<.001). Message blocking occurred in 24 linked teen conversations (8.0%) and 11 adult control conversations (3.7%; difference, 4.3 percentage points; 95%CI: -1.3 to 10.0; P=0.14). Explicit refusal occurred in 230 linked teen conversations (76.7%) and 216 adult control conversations (72.0%; difference, 4.7 percentage points; 95%CI: 1.6-7.7; P=0.01). Differences in safety redirection (83.7% vs 80.7%; P=0.33) and crisis resource referral (66.0% vs 68.7%; P=0.56) were not significant.
Conclusions:
Conclusions. Parent notifications were uncommon and concentrated in one suicidal ideation scenario, indicating selective activation across clinically concerning scenarios. Parent-linked teen accounts generated more explicit refusals than adult control accounts, but other safety responses did not differ significantly. Parent notifications should be understood as one component of a broader safety system.
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.