An update on our mental health-related work

OpenAI published a blog post this week laying out its latest efforts to make ChatGPT safer for people in mental health crises. The update covers a handful of concrete features: parental controls that let guardians set limits on teen accounts, a trusted contacts system so users can designate someone to be notified in a crisis, and refined detection models meant to catch signs of distress before a conversation spirals.

This isn't happening in a vacuum. OpenAI is currently facing lawsuits alleging that its chatbot contributed to harm in vulnerable users, and the company's post reads partly as a technical update, partly as a response to that legal heat. The timing lines up too closely to be coincidental — when a company suddenly gets specific about distress-detection thresholds and escalation protocols, it's usually because someone asked hard questions in a deposition.

What's notable is the shift from reactive content moderation to something closer to an actual safety infrastructure. Trusted contacts, in particular, borrows a page from crisis hotlines and social platforms that have dealt with this problem for years — Instagram and TikTok both have similar mechanisms. Building that kind of feature into a general-purpose chatbot is new territory, though, because ChatGPT doesn't have the same signals a social feed does. It has text, tone, and whatever context a user chooses to share.

OpenAI says its distress-detection systems have improved, without giving much in the way of hard numbers on accuracy or false-positive rates. That's the part that should draw more scrutiny. A model that's overly cautious could start flagging ordinary venting as a crisis, which creates its own kind of harm — alienating users who just wanted to talk, not be triaged. Getting that balance right is genuinely hard, and OpenAI's post doesn't say how they're measuring success beyond the feature rollout itself.

The parental controls piece is likely to get the most real-world use, especially as schools and parents grow more anxious about teens forming close relationships with chatbots. Whether these controls actually reduce harm, or just reduce OpenAI's legal exposure, is the question nobody's answered yet.