mpathic

mpathic is a clinician-founded AI safety company that helps AI developers evaluate, stress-test and improve human-facing models using behavioral science methods. Its services center on expert-led red teaming to surface failure modes, misalignment, bias and risks that automated tests and synthetic-data evaluations miss, and on ground-truth benchmarking that measures model performance on nuanced, high-stakes human behaviors. The company emphasizes detection of responses that pose physical or psychological risk to vulnerable populations before deployment, and translates evaluation results into insights that inform training data curation and fine-tuning. An optional product, mpathic Studio, supports AI-assisted annotation, benchmarking and reinforcement learning on models and multi-modal data.

The company works with a large pool of mental health experts, doctors, clinicians and other safety specialists, alongside actors, writers and red-teamers who build reinforcement learning and evaluation environments. Its benchmarking methodology compares model predictions to gold-standard human judgments: coders establish interrater reliability before creating ground-truth datasets, and models are scored on absolute agreement, precision, recall and F1 across domains including motivational interviewing, empathy and soft skills in medical settings, clinical risk and misconduct, and patient adherence. Published research themes include the mPACT benchmark family (a suicide benchmark, mPACT-S-v1.0, and an eating disorders benchmark, mPACT-ED-v1.0), fidelity and skill evaluation in clinical settings, safety and cultural responsiveness, LLM summarization for counseling, and multimodal applications such as robotic surgery communication and speaker attribution in clinical conversations.

Earlier descriptions of the company characterized it as enterprise software analyzing conversations across text, email and audio to flag potential misunderstandings and enable real-time adjustment — described by its CEO as "Grammarly for empathy" — with proprietary machine learning models identifying more than 100 conversational behaviors, and applications in clinical trial fidelity monitoring, adverse-event review and FDA adherence in CNS trials.

Founding story

Founder and CEO Dr. Grin Lord is a psychologist-researcher who spent roughly 15 years studying the elements of conversation that build trust and empathy, working as a clinician-researcher at the University of Washington's Department of Psychiatry and Behavioral Sciences from 2007, and later serving as Chief Clinical Officer at Lyssn.io and Clinical Product Psychologist at Youper.ai. She has said she started mpathic to make empathy — which she defines as accurate understanding — accessible to anyone through AI-driven coaching and correction, and the mission expanded into life-science applications such as detecting patient side effects or suicidal ideation in clinical conversations. Before mpathic, Lord and company architect Nic Bertagnolli started Empathy Rocks, a gamified platform using empathic AI.

Business model

mpathic sells expert-led evaluation services and software to organizations building human-facing AI. Engagements include red teaming, benchmarking against expert-rated ground truth, and delivery of model-ready insights for training-data curation and fine-tuning; customers can optionally use mpathic Studio for AI-assisted annotation, reinforcement learning and benchmarking. Earlier positioning described an enterprise conversation analytics and coaching platform plus an API for developers building emotion-AI applications, applied to clinical trial fidelity monitoring and medical monitor augmentation.

Revenue derives from paid engagements with AI builders for red teaming, benchmarking and evaluation, and from use of the mpathic Studio annotation and benchmarking product; earlier, from enterprise conversation analytics software and an API for emotion-AI applications.

Traction

As of June 2022 the company had 12 employees and had raised $5.2 million in total. It reports working with thousands of mental health experts, doctors, clinicians and other safety experts, and cites a customer in clinical trials that previously spent eight hours a day reviewing videos for adverse events. Public endorsements come from academic clinicians at Stanford, UC Berkeley and Washington University School of Medicine, and a clinical lead for mental health evaluation at a Fortune 100 AI lab is quoted as a client.

Latest developments

mpathic has published the mPACT benchmark family, including mPACT-S-v1.0 for suicide-related model behavior and mPACT-ED-v1.0 for eating disorders, and promotes expert-led red teaming and ground-truth benchmarking services alongside mpathic Studio for AI-assisted annotation and reinforcement learning. The executive team includes CEO and founder Dr. Grin Lord, co-founder and Chief Business Officer Dr. Danielle Schlosser, Chief Science Officer Dr. Alison Cerezo, COO and Chief Legal Officer Megan Fouty, CMO Rebekah Bastian, VP of Product Success Dr. Amber Jolley-Paige, and Chief Technology and Information Security Officer Brian Williams.