Red teaming with faith leaders: expanding digital safety and accountability to frontiers of care

Red teaming is a security exercise where invited testers simulate attacks to identify vulnerabilities in technical systems. It has become central to safety and accountability in end-user facing LLMs. Increasingly, red teaming has engaged outside experts to surface contextually specific vulnerabilities beyond cybersecurity domain expertise (i.e., ensuring LLMs cannot provide instructions for creating chemical weapons). Increasing focus on psychosocial harms has led to the engagement of mental health professionals. But this may be culturally-specific and potentially exclusionary: a majority of the world's population leverages non-psychological community supports during crisis. Faith leaders are one common example. With the growing use of LLMs for spiritual guidance and emergence of religious psychosis, their expertise is becoming newly salient. As an exploration of incorporating their expertise, we held a workshop with 11 faith leaders from Abrahamic traditions (Christian, Jewish, and Islamic) red teaming ChatGPT-5, complemented with results from a computational audit. We demonstrate how faith leaders challenge and extend cybersecurity epistemologies of scenario planning and testing, while still rigorously surfacing vulnerabilities. We find vulnerabilities and subsequent harms surfaced by faith leaders move beyond the individual to communitarian lenses, with novel mitigation implications. We end with practical and methodological guidance for how faith leaders, and other community leaders, can be trained and brought into digital safety work and accountability measures like red teaming to expand who is seen as a valid expert.