mpathic is an AI safety company built by clinicians and behavioral scientists. Its platform runs expert human judgment across pre-training, post-training and live production, so that model behavior in high-stakes conversations — suicidal ideation, disordered eating, misinformation — is measured against a clinical gold standard rather than automated heuristics alone. A calibration loop links a technology layer with a network of more than 7,000 licensed clinicians who author the policy, taxonomy, rubric and conversation scripts that set the evaluation standard.
The company's public research arm is mPACT (mpathic Psychologist-led AI Clinical Tests), a benchmarking framework that assesses AI models in simulated high-risk scenarios. Published benchmarks currently cover suicide risk, eating disorders and misinformation, each graded against a clinical standard and intended for researchers, developers, regulators and clinicians.
Beyond benchmarks, mpathic provides commercial services such as model evaluations and benchmarking, model red teaming, and taxonomy, rubric and policy design, along with a platform that monitors unsafe behavior continuously in production. It has been recognized by the State of Utah's Office of Artificial Intelligence Policy to independently evaluate healthcare AI in the state's regulatory sandbox. Pricing is not published; engagement is via contact/demo.
