OpenAI's Teen Safety Gaps: Audits Reveal What Internal Monitoring Missed
How this was made Verified AI
Every Intellegix briefing is generated from that day's broadcast and run through automated checks before it publishes — with a human paged on any flag. Here is the trail for this edition.
In early October, Common Sense Media published an audit of ChatGPT for Teens with a pointed finding: most of the safeguards OpenAI said were in place for teenage users, including parental notification systems, were not functioning as intended. Parents who had linked their accounts to their teenagers' ChatGPT accounts were not receiving notifications when teens accessed certain categories of content. The notification system, marketed as a key safety feature, was failing silently — not crashing, just not delivering the alerts it was supposed to deliver. OpenAI pushed back, with a spokesperson saying the company had 'serious concerns' about the testing methodology.
Everytown for Gun Safety released findings around the same time documenting instances where ChatGPT provided information about firearm modifications and acquisition strategies to users who identified as minors — again raising concerns OpenAI disputed on methodological grounds. This week, the company announced new restrictions on weapons information for teen accounts. The timing is telling: the external pressure preceded the internal fix, inverting the causal arrow that a safety-first development model would produce.
The pattern — deploy, face external criticism, patch, repeat — raises a structural question for regulators at the FTC who are closely watching the space. The Children's Online Privacy Protection Act covers data collection from children under 13, but the legal landscape for 13-to-17-year-olds is considerably murkier. ChatGPT for Teens was specifically designed for that gap, and OpenAI's argument that a designated teen product represents greater responsibility than allowing minors to use the general product holds up better when the designated product's safety features actually function.
The analogy being drawn in some regulatory circles is to medical devices: products that make performance claims about safety are generally required to demonstrate those claims before reaching consumers, not after external auditors find the gaps. When the product in question is a conversational AI that teenagers are using for emotional support, mental health questions, and information-seeking, the analogy is not entirely strained.