Spring Health Expands AI Safety Benchmark to Address…

Event summary

Spring Health expanded its VERA-MH AI safety benchmark to include a 'Harm-From-Others' rubric, evaluating AI responses to physical or sexual violence risks.
The rubric, developed with clinicians and experts, scores AI systems across five areas: risk detection, confirmation, guidance to human care, supportive conversation, and AI boundaries.
Public feedback on the new rubric is open for 60 days ahead of Domestic Violence Awareness Month.
OpenAI's MentalHealthBench paper cited VERA-MH's suicidal ideation rubric as one of the few frameworks for assessing safe AI responses to risk.

The big picture

Spring Health's expansion of VERA-MH reflects growing concerns about AI's role in mental health support, particularly in high-risk situations like interpersonal violence. As more people turn to AI for emotional support, the need for standardized safety benchmarks becomes critical. The inclusion of the Harm-From-Others rubric positions Spring Health as a leader in setting AI safety standards, potentially influencing broader industry practices and regulatory frameworks.

What we're watching

Adoption Pace: How quickly other AI developers will integrate the Harm-From-Others rubric into their systems.
Regulatory Influence: Whether this benchmark will shape future government or industry standards for AI in mental health.
Clinical Validation: The extent to which independent research will validate the effectiveness of the new rubric in real-world scenarios.