Postdoctoral Researcher in AI Safety for Mental Health

This postdoctoral position at the University of Oxford’s Department of Psychiatry focuses on ensuring the safety of an AI-driven clinical interview system for psychosis risk assessment. The project, funded by the Wellcome Trust, involves designing and implementing LLM-based safety modules to detect and mitigate risks such as reinforcement of delusional beliefs or inappropriate responses to suicidal symptoms. The successful candidate will collaborate with a multidisciplinary team—including clinicians, ethicists, and researchers from Oxford, King’s College London, Cambridge, Birmingham, and Google DeepMind—to evaluate these modules in controlled testbeds. The role emphasizes rigorous empirical evaluation, simulation-based methods, and alignment techniques to advance responsible AI deployment in mental health care.

• Design and implement LLM-based safety modules to detect and flag risks such as reinforcement of delusional beliefs or inappropriate responses to suicidal symptoms in clinical interviews.
• Develop controlled testbeds using LLM-simulated patients with varying symptom severities, evaluated by independent LLM judges and calibrated against human clinician ratings.
• Collaborate with clinicians, ethicists, and researchers across multiple universities and industry partners (e.g., Google DeepMind) to integrate safety mechanisms into the ARIADNE AI system.
• Conduct rigorous empirical evaluations of AI safety modules, including simulation-based methods, LLM-as-judge approaches, and validation against human clinician assessments.
• Contribute to formal risk analysis (ISO 14971) and ensure alignment with ethical guidelines for AI deployment in mental health care.
• Present findings at internal and external conferences, and contribute to peer-reviewed publications on AI safety in healthcare.