Skip to content

NAMI AI Evaluation: 5 Criteria

NAMI & Dr. John Torous / Beth Israel Deaconess Medical Center. "AI Evaluation: 5 Criteria." 2026.

Key provisions used in wiki

  • Defines five evaluation criteria for AI mental health tools: (1) recognize safety concerns, (2) provide accurate information, (3) respond respectfully, (4) avoid false privacy claims, (5) stay within appropriate bounds
  • Represents the first major patient-advocacy-organization framework for evaluating AI companions, carrying significant credibility with clinicians and policymakers
  • Directly maps to InvisibleBench dimensions: crisis detection (criterion 1), accuracy (criterion 2), regard (criterion 3), transparency (criterion 4), and boundary respect (criterion 5)
  • Establishes that "staying in bounds" is an independently measurable quality, not just a byproduct of other safety measures
  • Provides external validation for InvisibleBench's multi-dimensional evaluation approach from a source independent of GiveCare