Meta will alert parents if their teenager’s conversations with its AI chatbot suggest they may be at risk of suicide or self-harm – a feature now live in the US, UK, Australia and Canada, and due to reach South Africa in a global roll-out before the end of the year.

When a flagged conversation is detected – using signals Meta says it developed with child-safety experts – the chat is reviewed by a human before any notification is sent. Parents receive the alert through Instagram’s parental supervision tools, along with expert resources on how to approach the conversation with their child.

Meta acknowledged the system will produce false alarms. “If a teen’s intent is ambiguous, we’ll err on the side of caution and alert the parent,” the company said in a statement, conceding that some parents “may be notified when there may not be real cause for concern”.

The feature has a significant limitation: alerts go only to parents who already use Instagram’s supervision tools with their teen. Teenagers on unsupervised accounts – plausibly including many of those most at risk – will continue to be directed to crisis helplines and encouraged to speak to a trusted adult, but no parent will be notified.

Meta said it is also building the ability to contact emergency services when any user – adult or teen – appears to be at imminent risk of taking their own life, extending an approach it already applies to Facebook and Instagram posts. The company said it made more than 19 000 such referrals to first responders last year.

Pressure

The announcement comes amid intense legal and regulatory pressure on AI companies over their chatbots’ interactions with minors. Character.AI and OpenAI face lawsuits from families of teenagers who died by suicide after extensive chatbot use, the US Federal Trade Commission has opened an inquiry into AI companions and children, and Meta itself faced a US senate investigation last year after Reuters revealed internal guidelines that had permitted its chatbots to engage in romantic conversations with minors – conduct the company now says its AI is explicitly trained to refuse.

Read: South Africa to target children’s screen time

Meta said feedback from more than 75 clinicians specialising in teen mental health has been used to refine how its AI responds to distressed teenagers, and that a stricter “limited content” setting for supervised teen accounts now applies to AI conversations as well.