OpenAI tests age prediction to lock down ChatGPT for teens

If you spill your deepest secrets to an AI, should it be locked in a digital vault like a doctor-patient conversation, or should the company monitor the chat to make sure you do not hurt yourself? For OpenAI, those two goals are now directly at war.

OpenAI published a detailed memo laying out how it plans to navigate the brewing conflict between teen safety, user freedom, and privacy.

Treating adults like adults

OpenAI's first big stance is simple: treat adult users like adults. As frontier models become more steerable, the company wants to get out of the way of grown-ups.

Under the new stance, ChatGPT will not initiate flirtatious talk by default, but if an adult asks for it, the model should play along. Even on heavy topics like suicide, if an adult user asks for help writing a fictional story depicting it, the model should help rather than shut the prompt down.

On privacy, OpenAI is making an aggressive push. It argued that conversations with AI are among the most sensitive accounts anyone will ever have, comparing them to discussions with a doctor or a lawyer. The company is actively lobbying policymakers to grant AI conversations legal privilege.

To back that up, OpenAI is building technical guardrails so that even its own employees cannot read your chats. Exceptions remain only for massive harms, such as imminent threats to life, violent plans against others, or massive cybersecurity incidents, which will still escalate to human review.

Guessing ages and calling parents

When it comes to minors, however, that playbook gets thrown out the window. OpenAI explicitly stated that for teens, safety comes before both privacy and freedom.

To enforce this, OpenAI is building an age-prediction system that analyzes how people interact with ChatGPT to guess their age. If the system is uncertain, it will play it safe and default the account into an under-18 experience. In some regions, users may even be asked to verify their age with a government ID.

For anyone tagged as a minor, the guardrails tighten hard. ChatGPT will be barred from flirtatious banter, and self-harm or suicide discussions are completely blocked, even in creative writing scenarios.

Most strikingly, if an under-18 user shows signs of suicidal ideation, OpenAI will attempt to alert the user's parents. If that fails and harm appears imminent, the company will reach out directly to emergency authorities. New parental controls are also rolling out alongside the system.

Why it matters

We think this is one of the smartest and most honest product choices OpenAI has made in months. Blanket safety filters that treat every user like a fragile child have long frustrated serious writers and developers. Acknowledging that adults deserve the freedom to explore mature themes while keeping guardrails around minors is the right philosophical divide.

That said, the execution carries massive friction. Age-prediction based on chat patterns sounds neat in theory, but behavioral heuristics are notoriously easy to trick and prone to false positives. If adult users constantly get downgraded to the kids' table or forced to hand over state IDs, privacy-conscious users will rebel.

For the broader AI ecosystem, OpenAI is setting a precedent that rivals will inevitably have to copy. Demanding doctor-level legal privilege for AI prompts redefines consumer data rights, while calling the cops on suicidal teens thrusts AI makers squarely into the role of emergency services. The days of treating an AI chatbot as a passive software utility are officially over.