Research

Setting the standard: developing a benchmark for safe and effective AI in mental health support

This project will develop a practical, evidence-informed benchmark to evaluate the safety, effectiveness, and ethical standards of large language model AIs designed for use in mental health and therapy. 

It will bring together experts in counselling, clinical and systemic psychotherapy, people with lived experience of therapy, and technical specialists to co-design assessment tools that reflect the knowledge, skills, and values expected of qualified therapists. Through regular collaborative workshops and stakeholder engagement, the team will develop a suite of diverse, realistic case materials and a digital ‘test bed’ for evaluating AI systems.

By piloting this benchmark with current AI models, the project aims to set clear expectations for responsible use, support safer deployment, and offer practical guidance for industry, regulators, and service providers. This work will contribute to building public trust and ensuring equity and quality in digital mental health innovation.

Team

Laura Vowels – Lecturer, University of Roehampton

Angela Loulopoulou – Principal Lecturer, University of Roehampton

Sarah Ingram – Head of Strategic Partnerships and Programmes, Tavistock Relationships 

Laura Stembridge – CEO, InsideOut 

Rachel Francois-Walcott – Trainee Clinical Psychologist, University of Surrey