Project summary
Fide AI (fideai.org) will run a small research-credit microgrant program for faith-facing AI evaluation. The goal is to help researchers, pastors, and builders run concrete experiments on how AI systems behave in contexts involving theology, moral reasoning, spiritual formation, religious education, and pastoral-adjacent care.

What are this project's goals? How will you achieve them?
The goal is to turn early interest in Fide AI into useful research outputs. Since launching When AI Is Your Pastor and FMG-Bench, people from research, ministry, and AI-building contexts have reached out wanting to contribute.

We will select a small number of scoped projects, give contributors AI token/API credits, pair them with expert feedback where needed, and ask each project to produce a concrete output: an evaluation brief, scenario set, prompt comparison, source-grounding test, red-team note, or similar artifact.

Likely research areas include pastoral escalation, theological disagreement, source citation reliability, overtrust, spiritual companion risks, human agency, and faith-facing red-team probes.

How will this funding be used?
Requested budget: $10,000.

• $8,000 for AI token/API credits distributed to selected contributors.

• $2,000 for expert human feedback and review from pastoral, theological, technical, or research reviewers.

A smaller version still works: at $5,000, we would use $4,000 for AI credits and $1,000 for expert review.

Who is on your team? What's your track record on similar projects?
The project is led by Alex Chao, founder of Fide AI. Alex has worked at leading AI institutions like Uber ATG (self-driving cars), Microsoft's Office of the CTO, and Bytedance.

Fide AI recently launched When AI Is Your Pastor: A Benchmark for LLM Theological Triage and Pastoral Guidance, introducing FMG-Bench. The first release evaluated 14 advanced models across 8,792 scored responses, using 120 base scenarios and perturbation variants. The public package includes a research companion page, standalone benchmark repository, Hugging Face dataset, result summaries, caveats, and citation materials.

Fide AI has also launched a public research site and research agenda for faith-facing AI evaluation, with calls for research across grounding, pastoral escalation, reviewer reliability, formation, authority boundaries, and deployment readiness.

What are the most likely causes and outcomes if this project fails?
The most likely failure modes are:

• Contributor projects are too broad and do not produce useful artifacts.

• API credits get spent on exploratory work without enough methodological discipline.

• Expert review becomes a bottleneck.

• Outputs are interesting but not decision-relevant for institutions or builders.

To reduce those risks, we will keep grants small, require narrow project scopes, prioritize projects with a clear evaluation question, and reserve part of the budget for expert review.

If the project fails, the likely outcome is still bounded: a few experiments produce limited or inconclusive results, and we learn which contributor workflows are not worth scaling. The downside is small because the grants are small and the work is exploratory by design.

How much money have you raised in the last 12 months, and from where?
Fide AI has not raised outside funding in the last 12 months.

The work so far has been founder-funded and unpaid. We have submitted larger grant applications, but as of this application we have not received grant funding.