Google has published an update on its mental health-related AI work, detailing changes to how its models handle crisis and wellbeing queries [source]. The company states it has expanded SafetyCore, its framework for detecting and responding to mental health content, across additional products including Gemini and Search.

According to the post, Google now routes users expressing suicidal ideation or self-harm intent to crisis resources in over 30 countries. The system displays localised helpline numbers and connects users to organisations such as the 988 Suicide and Crisis Lifeline in the United States. Google reports these interventions have been shown to more than 10 million people since 2023.

The update also describes new guardrails for therapeutic advice. Gemini models will now decline to provide clinical diagnoses or treatment plans, instead suggesting users consult licensed professionals. Google states this change addresses feedback from mental health organisations concerned about AI overreach in clinical contexts.

The company has introduced a "wellness check-in" feature in Search for users querying symptoms of depression or anxiety. The feature offers validated screening tools from the National Alliance on Mental Illness and links to telehealth providers. Google emphasises these tools are informational and not diagnostic.

The post does not specify whether these changes apply retroactively to existing Gemini deployments or only to future releases. It also does not detail how SafetyCore distinguishes between users seeking information about mental health conditions and those in acute crisis. Google states it consulted with external mental health experts during development but does not name the organisations involved.