The Case for
JabezVault

The Premise

Modern large language models are trained on the open web. They do not distinguish between an authenticated primary manuscript and a paraphrased derivative scraped from a blog post. They generate confident, fluent, scripture-adjacent text at unlimited scale — and that text now circulates back into the same training corpus.

This is not a content problem. It is an infrastructure problem. There is no current system that guarantees the integrity of sacred primary sources at the scale and speed at which AI generates derivatives.

Documented Evidence

The following are real, recent examples in which AI systems produced false or fabricated scripture. Each is publicly reported.

1. ChatGPT fabricated entire Bible verses. A missions researcher writing for Missio Nexus prompted ChatGPT to "create a list of quotes from the Bible about truth and include the book, chapter, and verse." The model generated a verse attributed to John 5:5 that does not exist in Scripture; the actual John 5:5 reads, "one who was there had been an invalid for thirty-eight years." Four of the five verses in the list were fabricated. Source: Missio Nexus, "AI Hallucination, Chatbots, and the Truth of Holy Scripture."
2. ChatGPT generated a fake Jesus passage on demand. A widely shared incident, reported by The Christian Post and reviewed by multiple ministries, involved a user asking ChatGPT to write a "fake Bible passage" in the voice of Jesus. The model produced a fluent, scripture-styled paragraph beginning, "And a woman, whose heart was divided between spirit and body, came before him…" — a passage that does not exist in any manuscript tradition but reads, to an untrained ear, like authentic Gospel narrative. Source: Answers in Genesis coverage of the incident.
3. Bible Memory App introduced GenAI "Bible Intelligence" into a memorization tool. In July 2025, a long-trusted Scripture memorization app added a generative-AI feature marketed as "unlocking a deeper understanding of Scripture" by "connecting passages." Software engineer and longtime user Doug Smith publicly documented his decision to leave the app, warning that statistical word association produces "confident errors, made-up references, and hallucinations" inside a tool whose entire user base is built on the precision of the actual words of Scripture. Source: Doug Smith, "The Tragic AI-Powered Poisoning of The Bible Memory App."
4. ChatGPT confidently misinterprets and reverses interpretations of Scripture. A scholar writing for Christianity Today, adapted from The Biblical Mind, documented ChatGPT's tendency when interpreting the Sermon on the Mount to assert an interpretation, admit error when challenged, reverse course, and then quietly reassert the original claim — "metaphoriz(ing) and individualiz(ing) Scripture… without warrant, and often in direct contradiction to the text itself." Source: Christianity Today, "Misreading Scripture with Artificial Eyes."

The Technical Gap

Existing digital archives — from major library projects to denominational repositories — store scanned manuscripts. That is necessary, but it is not sufficient. None of them solve the problem AI creates downstream:

• • Scanned manuscripts are not cryptographically signed at the verse or token level.
• • Derivatives generated by AI are not flagged or fenced from the original.
• • There is no chain of custody between an authenticated primary text and any quotation, paraphrase, or summary later attributed to it.

In other words: existing archives store manuscripts. JabezVault verifies source integrity. The difference is the difference between a library and a vault.

Institutional Response

The concern is not fringe. Mainstream Christian and academic institutions have publicly raised the alarm:

• • Missio Nexus has published guidance to missionaries on the dangers of AI hallucination in scripture-related ministry.
• • Christianity Today has documented ChatGPT’s exegetical errors in detail.
• • Dallas Theological Seminary, Grove City College, and other institutions have begun publishing on the implications of generative AI for biblical scholarship and pastoral practice.
• • Mainstream secular outlets including NPR have reported on AI hallucination as a structural feature, not a bug, of large language models.

In every case, the diagnosis converges: AI generates confident, plausible, false content, and there is no infrastructure standing between that content and the trusting reader.

Solution Architecture

JabezVault provides sovereign infrastructure for sacred manuscripts with immutable provenance. The system is built on three architectural principles:

• • Authenticated primary sources. Manuscripts are ingested under documented chain of custody, with cryptographic signing at the manuscript and passage level.
• • Air-gapped sovereign hardware. The vault runs on dedicated, on-premises infrastructure — not hyperscaler cloud — isolated from public training pipelines.
• • Verified-only retrieval. Where AI assists scholarship inside the vault, every result is anchored to a signed primary source. There are no ungrounded generations.

This is the engineering answer to an engineering problem. The mission is theological. The solution is infrastructure.

Read more: Our Mission