Spring Health added a Harm From Others rubric to VERA-MH, its open source benchmark for mental health AI safety, and opened it for 60 days of public comment. The rubric evaluates how an AI system responds when an adult describes risk of physical or sexual violence from another person, scored across five areas using 100 personas, and is published in the VERA-MH repository.
Our readOpen safety criteria give buyers a common yardstick for mental health AI, including tools from other vendors. Buyers can use the draft rubric now to ask any vendor how its AI detects and escalates these situations.