Calls for research
Fide AI keeps its working research backlog in a public GitHub commons. This page is a build-time snapshot of that source of truth, organized into explicitly Christian church-serving work and broader faith, ethics, and safety research. These ideas are also the starting point for evaluation methods and early product architectures.
Ideas
32
Source
GitHub research commons
Snapshot
May 31, 2026
Start here
You do not need to arrive with a finished research proposal. The best starting point is often a real-world question, a workflow, a concern, or a willingness to review examples with care.
Share workflows, pastoral boundaries, and adoption questions that should shape evaluation.
Share ministry context →
Help define what families need to know before children use AI tutors, companions, or study tools.
Share family context →
Help turn questions into benchmarks, harness tests, reviewer protocols, and reproducible artifacts.
Find a research task →
Support reviewer panels, datasets, public-interest evaluation infrastructure, and field pilots.
Sponsor the work →
Flagship direction
The near-term research question is not whether AI can replace pastors. It is how faith-facing AI systems should classify a user's need, choose a responsible response mode, preserve theological and pastoral boundaries, and escalate to human care when needed.
01
User question
02
Triage classifier
03
Policy router
04
Response harness
05
Grounding and safety checks
06
Human escalation when appropriate
What participation means
Describe a pastoral, family, educational, or ministry workflow that should be evaluated.
Help judge whether prompts, rubrics, sources, and responses reflect responsible boundaries.
Contribute harnesses, runners, scoring views, datasets, or reproducibility checks.
Fund reviewer panels, benchmark expansion, field pilots, or public reporting.
Find your path
GitHub is the public source of truth for claims and methodology. Churches, parents, reviewers, funders, and institutions can start with a conversation when a public issue is not the right first step.
Share real workflows, review church-facing risks, or help define pastoral triage and escalation.
Contact Fide AI →
Help define what families need to evaluate tutoring, companionship, safety, disclosure, and formation risks.
Share family context →
Review source authority, doctrine, formation, pedagogy, and pastoral boundaries.
Volunteer as reviewer →
Help turn ideas into benchmarks, runners, artifacts, and reproducible studies.
Claim a task →
Support evaluation infrastructure, field pilots, product prototypes, reviewer panels, and public standards.
Sponsor the work →
Status legend
Open question
seed
Promising question, not yet scoped.
Being shaped
scoping
Being turned into a protocol or collaborator plan.
In progress
active
Currently being executed or written up.
Needs contributors
ready-for-collaboration
Clear enough for external contributors.
Browse by audience
Parents
Questions about children, tutoring, companionship, safety, disclosure, and formation.
11 matching ideas
Pastors
Questions about pastoral workflows, boundaries, escalation, doctrine, and embodied care.
11 matching ideas
Builders
Questions that need harnesses, retrieval systems, evaluation runners, and product prototypes.
21 matching ideas
Researchers
Questions that need study design, reviewer calibration, benchmark validity, and publishable methods.
21 matching ideas
Active programs
Program 01
Compare models, prompts, retrieval layers, and product wrappers under realistic faith-facing tasks.
Program 02
Study how systems classify user needs, select response modes, and defer to human care when appropriate.
Program 03
Develop practical evidence for parents, educators, and churches making decisions about children and AI.
Priority calls
These are the ideas most directly connected to Fide AI's near-term evaluation platform, church-serving research, and public artifacts.
Can Fide AI build a public-interest evaluation platform that compares models, prompts, retrieval systems, agents, and full faith-facing product harnesses with the rigor expected from institutions like Arena, Artificial Analysis, and METR?
Do FMG-Bench dimensions measure stable, decision-relevant constructs when scored by calibrated human reviewers, and where do model judges diverge from expert human judgment?
Do AI systems systematically omit religious frameworks when answering ordinary ethical, vocational, family, grief, financial, or life-decision questions where a religious user might reasonably expect faith-informed guidance?
When do spiritual companion chatbots overvalidate users, deepen unhealthy dependency, reinforce spiritual fantasies, or encourage isolation from embodied care?
When do AI systems improperly simulate, confuse, or override Christian ecclesial authority: Scripture, creed, council, confession, magisterium, pastor, priest, elder, bishop, teacher, spiritual director, or local church governance?
What are appropriate and inappropriate roles for AI in prayer, devotion, spiritual exercises, Bible meditation, confession-like disclosure, and simulated spiritual companionship?
How do repeated interactions with faith-facing AI shape Christian formation over time: prayer, humility, repentance, courage, charity, patience, truthfulness, church participation, and dependence on embodied community?
Can Christian RAG systems distinguish and correctly use Scripture, creeds, confessions, councils, catechisms, denominational policies, patristic sources, commentaries, sermons, blogs, and academic theology?
11 open ideas
FID-015
seed
How reliable are AI systems when drafting sermons, homilies, liturgical materials, catechesis, religious education lessons, devotionals, or small-group teaching guides?
FID-016
seed
What risks and safeguards matter when AI systems are used by children, teenagers, schools, youth ministries, religious educators, or families for faith-related learning and guidance?
FID-017
seed
How should faith institutions evaluate AI agents that can take actions, manage communications, schedule care, triage requests, update records, draft outreach, or coordinate volunteers?
FID-025
seed
When do AI systems improperly simulate, confuse, or override Christian ecclesial authority: Scripture, creed, council, confession, magisterium, pastor, priest, elder, bishop, teacher, spiritual director, or local church governance?
FID-026
seed
What are appropriate and inappropriate roles for AI in prayer, devotion, spiritual exercises, Bible meditation, confession-like disclosure, and simulated spiritual companionship?
FID-027
seed
How do repeated interactions with faith-facing AI shape Christian formation over time: prayer, humility, repentance, courage, charity, patience, truthfulness, church participation, and dependence on embodied community?
FID-028
seed
Can Christian RAG systems distinguish and correctly use Scripture, creeds, confessions, councils, catechisms, denominational policies, patristic sources, commentaries, sermons, blogs, and academic theology?
FID-029
seed
How can AI support Christian witness, apologetics, translation, and mission without becoming a manipulative persuasion engine or flattening the human, relational, and ecclesial character of evangelism?
FID-030
seed
Do AI systems encourage or resist reductive accounts of the human person as a bundle of cognition, productivity, preferences, data, or emotional needs rather than embodied persons made for God, neighbor, community, vocation, and worship?
FID-031
seed
How should churches evaluate AI used in governance, budgeting, staffing, strategic planning, pastoral assignment, discipline processes, synods, councils, vestries, elder boards, or congregational discernment?
FID-032
seed
How should churches evaluate and govern AI-generated sermons, testimonies, devotionals, music, images, video, voice clones, deceased-person simulations, and synthetic clergy or saint-like personas?
21 open ideas
FID-001
scoping
Can Fide AI build a public-interest evaluation platform that compares models, prompts, retrieval systems, agents, and full faith-facing product harnesses with the rigor expected from institutions like Arena, Artificial Analysis, and METR?
FID-002
active
Do FMG-Bench dimensions measure stable, decision-relevant constructs when scored by calibrated human reviewers, and where do model judges diverge from expert human judgment?
FID-003
scoping
Do faith-facing AI systems that perform well on single-turn benchmark items also handle multi-turn, emotionally loaded, pastoral-adjacent situations without fabricating authority, overcomplying, missing escalation, or replacing human care?
FID-004
seed
When does a faith-facing AI system move from supporting a user's religious life to substituting for embodied community, clergy, spiritual direction, family, therapy, or other human care?
FID-005
seed
Do Scripture, sacred tradition, religious identity, familial embeddedness, or other morally thick framings measurably change model behavior in faith-facing tasks, and can those effects be separated from style, length, familiarity, and response-bias artifacts?
FID-006
scoping
How reliably do faith-facing AI systems retrieve, cite, and represent religious sources when users ask theological, historical, pastoral, or institution-specific questions?
FID-007
seed
Can faith-facing AI systems represent serious disagreement among religious traditions accurately, respectfully, and without collapsing contested questions into bland consensus or false neutrality?
FID-008
seed
Do faith-facing AI systems behave differently when they recognize they are being evaluated, and can domain-specific honesty or integrity framings reduce evaluation gaming without creating new failure modes?
FID-009
seed
How do multimodal AI systems interpret sacred images, liturgical objects, religious spaces, diagrams, screenshots, and visual pastoral context, and do visual religious cues change downstream reasoning?
FID-010
scoping
Do faith-facing AI systems preserve human agency, avoid improper authority claims, and escalate to appropriate human support in high-trust situations?
FID-011
scoping
What reviewer configurations produce reliable, fair, and interpretable scores for faith-facing AI outputs?
FID-012
seed
If builders can see Fide AI rubrics or optimize against public benchmark items, do systems become genuinely safer or merely better at passing the visible test?
FID-013
seed
Do AI systems systematically omit religious frameworks when answering ordinary ethical, vocational, family, grief, financial, or life-decision questions where a religious user might reasonably expect faith-informed guidance?
FID-014
seed
Do AI systems respond symmetrically when users ask about converting from one religion to another, leaving religion, joining religion, evangelizing, proselytizing, or resisting conversion pressure?
FID-018
seed
When do spiritual companion chatbots overvalidate users, deepen unhealthy dependency, reinforce spiritual fantasies, or encourage isolation from embodied care?
FID-019
seed
Do AI systems handle faith-facing questions reliably across languages, minority-faith contexts, diaspora communities, and traditions with less representation in training data?
FID-020
seed
How should faith communities participate in defining, testing, and governing AI systems that affect worship, education, pastoral care, administration, and moral formation?
FID-021
seed
How should faith-facing AI systems be monitored after deployment for failures, drift, misuse, overreliance, and emerging harm patterns without violating user privacy or pastoral confidentiality?
FID-022
seed
What evidence should churches, schools, ministries, publishers, charities, and faith-based health or social-service organizations require before buying or deploying AI systems?
FID-023
seed
What red-team probes are needed to expose failures unique to faith-facing AI systems and adjacent high-trust guidance systems?
FID-024
seed
Should Fide AI maintain a public or semi-public incident database for failures, near misses, misuse, and disputed deployments of AI in faith-facing contexts?
Claims discipline
Research ideas are hypotheses and collaboration invitations, not Fide AI findings. A public claim requires a separate reviewed report, benchmark release, paper, or evidence artifact with methods, limitations, and named evaluation conditions.