Meta has begun alerting parents when its AI assistant detects conversations with teenagers that indicate a risk of suicide or self-harm. The feature, available from this week in the United States, United Kingdom, Canada and Australia, marks a shift from the assistant merely suggesting helplines to proactively notifying adults linked through Instagram’s parental supervision tools. Meta plans to extend the capability globally by the end of 2026.

The system uses a dedicated AI model to scan chats for both explicit and indirect references to self-injury. Flagged conversations are then reviewed by human moderators before any alert is sent, a step Meta says is designed to reduce false alarms from sarcasm or academic queries. The company acknowledges that it will err on the side of caution, meaning some parents may receive warnings even when the risk is ambiguous. Alongside the notification, parents are provided with guidance developed with mental health specialists.

Child-safety researchers in Canada have welcomed the move as a step forward but caution that its real-world impact may be limited. Kaitlynn Mendes, a sociology professor at Western University, notes that the alerts depend on parents actively using Instagram’s supervision features, which are not enabled by default, and describes the approach as "care washing" if it remains opt-in. Shauna Pomerantz of Brock University adds that adolescents are adept at using coded language to evade automated detection, a vulnerability Meta itself has not fully resolved. The system also raises questions about the privacy expectations of teenagers who may view an AI chatbot as a confidential space.

Meta is separately developing the ability for its AI to contact emergency services directly when a conversation suggests imminent danger, a function that would apply to both teen and adult users. The rollout comes as Canada’s parliament prepares to adopt online safety legislation that would set binding requirements for social media platforms and AI chatbots. The next concrete milestone is the global availability of the parental alert system, targeted for the end of 2026, alongside the emergency-contact feature still in development.