New research demonstrates how AI models can be tested to ensure they don’t cause harm by revealing anonymized patient health data.