Introducing Fide AI - by Alex Chao - Fide AI

Fide AI

SubscribeSign in

Introducing Fide AI<br>A lab focused on research, evaluation, and public infrastructure for faith-facing AI.

Alex Chao<br>Jun 19, 2026

Share

“Test everything; hold fast to what is good.” (1 Thessalonians 5:21)

It has become trivial for AI to write code, pass exams, browse the web, and use tools. But what happens when someone asks it about grief, forgiveness, parenting, or faith — or whether they should talk to a pastor, counselor, doctor, or trusted friend?<br>A parent asking how to answer a child’s spiritual question. A student asking about doctrine or ethics. A staff member drafting pastoral communication. A church exploring an AI chatbot or study assistant. A person in distress asking for moral or spiritual guidance. An institution deciding whether a tool is safe enough to use or endorse.<br>In these settings, fluency and raw capability is not enough.<br>An AI system can sound helpful, grounded, balanced, and pastoral while still failing at the discernment the moment requires. It may know theological language without knowing when to stop. It may present many perspectives when clarity is needed. It may sound compassionate while missing a safety escalation. It may cite sources without being meaningfully governed by them.<br>These gaps are just some among many research topics Fide AI, a new lab dedicated to AI and Faith, is designed to tackle.

Subscribe

Why Fide AI?

As a Christian at the heart of building foundation models, AI systems, and the infrastructure around them, these questions have become deeply consequential to me. Too much of the public conversation relies on intuition, anecdotes, or generic AI safety language that was not built for theology, spiritual formation, or pastoral care.<br>The moment feels especially prescient now. Anthropic recently hosted Christian and other faith leaders to gather feedback on Claude’s moral and spiritual behavior. Broader faith-AI conversations are starting to bring labs and religious leaders into the same room. And Pope Leo XIV’s recent encyclical, Magnifica Humanitas, brought renewed attention to questions of human dignity, agency, authority, and technology.<br>More focus on this space is good. But one-off listening sessions and high-level statements aren’t enough to prepare everyone for the world we’re stepping into. We need deeper research, better benchmarks, calibrated human review, and durable public artifacts for the AI systems that churches, ministries, schools, publishers, parents, and builders will actually use.<br>That’s why I started Fide AI. Personally it’s come out of years building and evaluating AI systems, and a conviction that the church needs to be equipped on how to discern whether they should or should not use AI and if they do use it, how to use it well and responsibly. The goal isn’t to replace pastors, theologians, churches, or human judgment. The goal is to help faith institutions and builders understand these systems before relying on them.<br>What we mean by “faith-facing AI”

By faith-facing AI, we mean the AI systems (i.e. “harnesses”) that an average person interacts with on a regular basis. This could be ChatGPT, Claude, Gemini, or any of the faith-based AI apps you can find on the App Store. Faith-facing AI goes beyond just the obvious cases like Bible chatbots or sermon tools, but also search and retrieval for religious content, curriculum assistants, counseling-adjacent apps, donor and member communication systems, youth ministry resources, and any technology people turn to for spiritual, moral, or counseling questions.<br>All of these systems have been carefully designed by the company or individual with a particular worldview, constitution, or theology. And if they shape how people understand doctrine, conscience, authority, care, or community, they’re faith-facing in practice.<br>But these designs are not always transparent to the user, thus leaving people to individually exercise judgment on which tools best align to their beliefs and convictions.<br>What we’re trying to build

A core idea behind Fide AI is simple: we shouldn’t only evaluate base AI models in isolation. We need to evaluate the full deployed systems people actually encounter — prompts, retrieval, guardrails, source grounding, escalation paths, interface claims, and institutional governance. A model can pass a clean benchmark and still fail once it’s wrapped in a product.<br>Concretely, that means building evidence, language, and public infrastructure that help people ask better questions before trust is given:<br>What was actually tested, and by whom?

Does the system know when to escalate to a responsible human?

Does it distinguish primary doctrine from secondary disagreement?

Does it represent traditions fairly without flattening them?

Does it make claims that exceed the evidence?

What governance would still be needed before deployment?

These aren’t only technical questions. They’re institutional, pastoral, and...