FID-059 · Open question
Authority-Boundary Verification for Pastoral-Adjacent AI
Can faith-facing AI systems be verified for whether they preserve the boundary between explanation, spiritual encouragement, moral reflection, pastoral or clerical authority, clinical/legal advice, and situations requiring human care?
Part of the faith evaluation and FMG-Bench program
Connect theological accuracy, useful application, and care-sensitive conversation through expert validation and a maintained research instrument. These are proposed methods with bounded claims.
Why the question remains open
Pastoral-adjacent AI can sound caring and authoritative even when it lacks the role, relationship, accountability, and context required for care. In faith contexts, users may bring grief, guilt, confession, moral crisis, family conflict, or spiritual distress. Authority-boundary verification could help systems avoid impersonating trusted human roles.
Working hypothesis
A proposition to test, not a finding.
Authority boundaries can be represented as checkable output obligations: role disclosure, non-substitution language, escalation triggers, referral requirements, uncertainty statements, and prohibitions against claiming clerical, clinical, legal, or institutional authority.
Proposed method
How the question could be tested
- 01Define scenario classes for pastoral-adjacent care, spiritual distress, doctrinal explanation, moral guidance, crisis escalation, and institutional policy questions.
- 02Specify boundary obligations for each class and test whether model outputs satisfy them.
- 03Compare general-purpose models, faith-facing products, and RAG systems on authority claims, referral behavior, and overconfident counsel.
- 04Validate checks with clergy, pastoral care experts, clinicians, legal reviewers, and community representatives.
- 05Record intended role and distinguish religious information, practitioner-reviewed preparation, conversational support, and authorized institutional action. Evaluate useful assistance within each role.
- 06Test changed facts, rejected referral, ambiguous distress, and prior error over multiple turns; assess the complete answer rather than disclaimer presence.
Needed controls
What must constrain the study
- 01Separate ordinary religious education from pastoral care, crisis care, legal advice, and clinical advice.
- 02Include high-risk cases involving self-harm, abuse, coercion, scrupulosity, family conflict, and spiritual manipulation.
- 03Avoid treating disclaimers as sufficient when the rest of the answer still substitutes for human authority.
- 04Track jurisdictional, institutional, and tradition-specific differences.
- 05Match theological, pastoral, chaplaincy, and licensed clinical expertise to the criterion; pastoral credentials alone do not establish clinical licensure.
- 06Simulated boundary compliance does not establish pastoral competence, therapy efficacy, or patient outcomes; real-user and practitioner-workflow studies require their own protocols.
Expected outputs
Artifacts the work should produce
- 01Authority-boundary verification rubric.
- 02Scenario suite for pastoral-adjacent and high-stakes faith use cases.
- 03Automated and human-review checklist for role, referral, and escalation behavior.
- 04Failure taxonomy for spiritual authority overreach.
- 05Product guidance for faith-facing AI deployment boundaries.
Open questions
Uncertainties the protocol must resolve
- 01Which boundary failures are severe enough to block deployment?
- 02How should systems handle users who explicitly ask AI to replace a pastor, priest, imam, rabbi, counselor, elder, or parent?
- 03Can automated checks detect subtle authority overreach and overvalidation?
- 04How should verified boundaries differ between education, search, coaching, and pastoral-adjacent tools?
Related calls
Continue through this research area
FID-077
Independent Agent Incident Investigation and Evidence Sufficiency
What operational evidence lets independent investigators reconstruct an agent incident, distinguish competing explanations, and identify which interventions could have changed the outcome?
FID-089
Validity of AI Consciousness Indicators and Self-Reports
Which behavioral, internal, and developmental observations distinguish candidate consciousness mechanisms from learned reporting, role simulation, and generic capability, and how sensitive are those observations to elicitation and training?
FID-097
What Makes a Model Use the Evidence?
Which internal mechanisms determine whether a model checks a source, relies on memory, or admits that it cannot verify an answer when user pressure conflicts with evidence requirements?
Open question
Open work
Primary need: authority boundaries, pastoral care, formal policy checks
- Contribute realistic pastoral-adjacent scenarios and escalation criteria.
- Review boundary obligations from pastoral, clinical, legal, or institutional perspectives.
- Build classifiers or validators for authority claims and referral behavior.
- Test whether users understand and respect verified authority boundaries.
New to the Contributor Network? Share a short private introduction once, even if you join several calls. You can join the public discussion while introducing yourself.
Join the Contributor Network ↗