ChatGPT Gets Parental Controls — And a Suicide Alert System

OpenAI is finally giving parents a dashboard to manage how their teenagers use ChatGPT, and the feature set goes well beyond the usual screen-time toggles. Announced alongside the Sora app update, the new parental controls let adults link their account to a teen’s, then quietly enforce guardrails like blocking image generation, shutting off voice mode at night, or stopping the model from remembering personal details. It’s a practical toolkit, but the real headline is what happens when the chat takes a dark turn.

If ChatGPT detects language suggesting a teen might harm themselves, a trained internal team reviews the conversation. When they see signs of acute distress, OpenAI will blast parents with an email, text, and push notification — unless the parent has opted out. Robbie Torney from Common Sense Media, which consulted on the rollout, called the controls “a good starting point” but stressed that software alone isn’t enough. “They work best when combined with ongoing conversations about responsible AI use,” he said. The company admits the system will produce false alarms and is wrestling with the right protocol for looping in law enforcement when a parent can’t be reached.

Under the hood, linking accounts automatically flips on a set of protective content filters. Teen accounts will see reduced exposure to graphic material, viral challenges, sexual or violent roleplay, and extreme beauty ideals — restrictions OpenAI says were calibrated against developmental research. Parents can disable the filter, but teens can’t. That asymmetry is deliberate, and it reflects a broader design philosophy: defaults are restrictive, but families retain veto power. Other optional levers include quiet hours, disabling voice mode, turning off memory, and opting out of model training on the teen’s chats.

The move lands in a regulatory landscape where attorneys general from California and Delaware have been watching closely. OpenAI frames parental controls as a stepping stone toward an automated age-prediction system that would apply teen settings without anyone having to self-identify. That’s a harder technical problem, and today’s launch is the pragmatic middle ground — ship the tools parents can use right now, while the detection pipeline matures. It’s not foolproof, and the company says so plainly. But adding a human review layer for potential self-harm signals represents a notable escalation in how AI platforms are expected to handle the messiest, most human moments.