OpenAI linked parent and teen accounts with distress alerts

What happened

OpenAI began describing its teen-safety plans in a 16 September 2025 post, Building towards age prediction, which said the company was 'building toward a long-term system to understand whether someone is over or under 18.' Two weeks later, on 29 September 2025, a second post, Introducing parental controls, announced that the feature was live to all ChatGPT users that day: a parent or guardian can send a linked-account invite to a teen, and once accepted, manage settings for that teen's account from a control page.

What the documents show

The parental-controls post states that linked teen accounts automatically receive 'reduced graphic content, viral challenges, sexual, romantic or violent roleplay, and extreme beauty ideals,' a default parents can turn off but teens cannot. It also describes a notification system: if the company's systems detect 'potential signs that a teen might be thinking about harming themselves,' a trained review team assesses the case, and in situations of 'acute distress' the company will contact parents by email, text and push alert unless they opted out, with law-enforcement contact reserved for an imminent threat the company cannot resolve by reaching a parent. The company states it worked with Common Sense Media and the Attorneys General of California and Delaware in shaping the approach.

The mechanism

Two mechanisms operate together. Content safeguards act at the point of generation, filtering categories of output for any account linked as a teen, independent of what the teen types. The distress-notification system instead relies on detection: pattern recognition across a teen's messages, escalated to human review, then to a family contact — a probabilistic judgement rather than a fixed rule, which the company itself describes as capable of raising 'an alarm when there isn't real danger.'

What it leaves open

Neither post states the detection system's accuracy, how many false alerts it has produced, or what a teen sees when a notification is triggered without their initiating it. The age-prediction system that would apply these protections without a parent's action is, as of these posts, still described as forthcoming rather than shipped. Editorially, the gap between promising to predict age and requiring a parent to first link the account is the practical limit of this record as of its retrieval date.

• What happens to a teen's ability to unlink their account once a parent has set restrictions?
• How does the company distinguish acute distress from a teen's ordinary difficult conversation?
• Until age prediction ships, what protects a teen using an unlinked, adult-registered account?

The two posts together show a company sequencing its response: a stated future capability first, then a narrower, opt-in tool that depends on a parent acting, which is a meaningfully different safeguard than one applied automatically to every under-18 user.