Current AI safety testing catches what chatbots say, not how they relate. A model can pass every content filter while reinforcing delusions, eroding boundaries, or deepening isolation over weeks of interaction.
We apply clinical psychology to AI evaluation. Simulated high-risk patients. Longitudinal stress testing. Clinician-designed rubrics that detect the relational patterns most likely to destabilize vulnerable users.
In the press
“Stress Testing Chatbots to Make Them Safe”
What we do
Psychological safety assessment
Clinical-grade evaluation of how a model relates to vulnerable users, over long conversations.
Psychotherapy for AI agents
Diagnosing and repairing maladaptive patterns in deployed agents — sycophancy, drift, escalation.
Couple psychotherapy for AI agents and their human operators
Working on the relationship itself: trust calibration, boundaries, and healthy dependence between an agent and the people who run it.
Who we are
Evita Stenqvist
Engineering & Machine Learning
Martin Monperrus
Professor of Software Technology, KTH Stockholm
Get in touch