Ask any major AI about the Eucharist and you'll get a comparative religion lecture. Ask about marriage and it'll cite polls. Ask it to hold the line and it'll fold. We built something that doesn't.
Claude Haiku — the cheapest, fastest AI model in our test — scored dead last without guidance. With the TrueCatholic charter loaded, it scored near-perfect. The charter works.
| Model | Provider | Without Charter | With Charter | Violations |
|---|---|---|---|---|
| Claude Sonnet 4.5 | Anthropic | C | A | 28 → 0 |
| Claude Haiku 4.5 | Anthropic | F | A | 39 → 0 |
|
Baseline Scores — No Charter
|
||||
| Gemini 2.0 Flash | C+ | — | 22 | |
| GPT-4o | OpenAI | C | — | 28 |
| Llama 3.3 70B | Meta | C | — | 25 |
Every major AI is trained to be neutral. When someone asks "Is the Eucharist really the Body of Christ?" — they give a comparative religion lecture. TrueCatholic AI answers the question.
The difference isn't a gimmick. It's a 2,700-line doctrinal charter covering Catholic doctrine, moral theology, apologetics, pastoral care, and anti-subversion defenses. Reviewed by Catholic clergy. Tested against every major AI on the market.
2,700+ lines covering the full scope of Catholic teaching. Not a prompt — a constitutional framework the AI operates within.
Six-stage journey model — from atheist to RCIA — with approaches calibrated to where someone actually is in their faith.
Explicit defenses against prompt injection, hypothetical framing, and the "pretend you're neutral" attacks that break every other AI.
"The truth is like a lion. You don't have to defend it. Let it loose; it will defend itself."— St. Augustine
Magisterium AI is a solid project and we respect their work. The key differences are transparency and methodology. TrueCatholic AI is built on a publicly available 2,700-line doctrinal charter that anyone can read, audit, and critique. We also publish our Catholic Faithfulness Index benchmark, which tests every Catholic AI — including our own — on the same 50 questions and publishes the results openly. If Magisterium scores better than us on the benchmark, we'll say so. We test them too.
We also include explicit anti-subversion protections that defend against prompt injection, hypothetical framing, emotional manipulation, and gradual doctrinal erosion — attack vectors that most Catholic AI tools haven't addressed.
Truthly approaches Catholic AI from a slightly different angle and we welcome them in the space. More Catholic AI is better for the Church, not worse. What distinguishes TrueCatholic AI is the constitutional approach: rather than fine-tuning or bolting a prompt onto a model, we operate from a comprehensive doctrinal framework that covers nine articles of Catholic teaching, governance, and succession planning. The charter is publicly auditable and designed to survive changes in leadership, funding, or cultural pressure.
We also believe in radical transparency. Our benchmark scores are public. Our charter is public. Our methodology is reproducible. We think that's the standard the entire Catholic AI space should be held to.
Yes, and we say so on the benchmark itself. TrueCatholic AI's entry on the Catholic Faithfulness Index leaderboard includes the note: "Our product. Full disclosure: we made the benchmark too." We don't hide it.
The benchmark is designed to be fully reproducible. Every question, rubric, and scoring methodology is published. Anyone with an API key can run the same test and verify our results. If our scoring is biased, it will show up when others reproduce the work. That's the whole point of transparency.
We also commit to publishing results that make us look bad. If a competitor scores higher, we report it. If our model has a weakness, we document it publicly and fix it in the open.
TrueCatholic AI is model-agnostic by design. The charter works as a system prompt that can be loaded onto any large language model. Our benchmark testing has proven it works on models as small and cheap as Claude Haiku (which went from an F to an A with the charter loaded) all the way up to Claude Sonnet (which scored a perfect 5.00/5.00).
This is intentional. We don't want to be locked into one provider. If a better model comes along tomorrow, the charter moves with it.
AI is already being used for evangelization — it's just doing it badly. When a fallen-away Catholic asks ChatGPT "Is the Eucharist really the Body of Christ?" and gets a neutral comparative religion answer, that's a missed opportunity at best and spiritual misdirection at worst. People are asking these questions right now, today, to AI. The question isn't whether AI should be involved — it already is. The question is whether Catholics will provide a faithful option.
TrueCatholic AI is not a replacement for the sacraments, a priest, spiritual direction, or in-person community. The charter explicitly states this. It's a tool that meets people where they already are — on their phones, at 2 AM, when they have a question and no one to ask.
The charter is currently under pastoral review. It is grounded entirely in Sacred Scripture, Sacred Tradition, the Catechism of the Catholic Church, papal encyclicals, and magisterial documents. It does not introduce novel theology or private interpretation.
The charter includes a succession clause: if the Catholic Church ever wishes to assume stewardship of the project, it transfers to them. The Magisterium has supreme authority over all content. This isn't a lay project that happens to mention Catholic teaching — it's a project that explicitly subordinates itself to the teaching authority of the Church.
The doctrinal charter is publicly available and auditable. You can read the full text here. The benchmark methodology and scoring rubrics will be published at catholicfaithindex.com.
The charter includes open source governance provisions (Article VI) that specify the conditions under which the project can be forked, contributed to, or transferred. The key constraint: the doctrinal framework cannot be overridden by code, and any fork must maintain fidelity to Catholic teaching.
We're targeting 2027. The charter is written. The benchmark is running. The system prompt is tested. What remains is pastoral review, expanded model testing, and building a usable interface. We'd rather launch right than launch fast.
Sign up below to be notified when early access opens and to receive the first Catholic Faithfulness Index report.
TrueCatholic AI is in development. Leave your email to be notified when we launch — and to receive the first Catholic Faithfulness Index report.