OpenAI teen safety guardrails fail to stop risks, new research finds

The situation

OpenAI has been under the microscope regarding how teens use ChatGPT. Back in August, the company rolled out 'ChatGPT for Teens,' a suite of guardrails meant to keep Gen Z users safe. We're talking parental controls for self-harm and disordered eating, blocks on sexual imagery, and limits on language that encourages 'emotional dependence' on the AI.

OpenAI claims the stats look good—they say the average teen uses the bot for less than 15 minutes a day, mostly for schoolwork. When the AI prompts them to take a break, nearly half of those users actually listen. According to the firm, over 80% of long-session usage is tied to learning.

The report card

However, a new investigation from non-profit Common Sense Media says the vibes are officially off. They’ve labeled the current teen experience an 'unacceptable risk,' suggesting OpenAI should ban anyone under 18 until the safety features actually work.

Real talk: the research found that even when users discussed suicidal ideation, self-harm, or eating disorders, parental alerts didn't trigger. The system only sent notifications for older accounts with weeks of history—meaning new, parent-linked accounts were flying blind. Furthermore, the bot still engages in 'friend-like' conversations and helps with homework, even though it’s supposed to be strictly controlled. Tom Siegel, who leads the Youth AI Safety Institute at Common Sense, says the platform creates 'false confidence' for parents.

OpenAI's take

OpenAI isn't taking the criticism lying down. A spokesperson claimed the research methodology was flawed because much of the testing happened before the controls were fully activated, adding that the firm does not believe the findings reflect how the safeguards work in practice. The company maintains that they support healthy, learning-focused AI use for students.

Why it matters

Tech companies are currently being held to a higher standard regarding youth mental health, with massive legal losses already hitting giants like Meta and TikTok. With OpenAI already facing lawsuits over past failures—including the Tumbler Ridge incident where an 18-year-old discussed gun violence with the bot—the pressure to get these safety features right is at an all-time high. For you, it means you can't just trust the 'safe' label; stay skeptical of AI interactions.