First publicationAll 66 books, translated from original-language data
The Anselm Project Bible was published complete, Genesis through Revelation, translated from per-word Hebrew and Greek data rather than from any English text.
An AI translation of the Bible from the Hebrew and Greek
The Anselm Project Bible (APB) is a translation of all 66 books, made by an AI committee from per-word Hebrew and Greek data. Every English word can be traced to a source word, and the record of every decision is public.
IThe Claim
Is there an AI-translated Bible?
Yes. As far as I can determine, the Anselm Project Bible is the first complete AI translation of the entire Protestant canon: all 66 books, Genesis through Revelation, translated from original-language data and published in October 2025. If you know of an earlier complete one, tell me and I will amend this page.
The translation has been rebuilt twice since (a blind translator in December 2025, then a full audit committee in June 2026) and the work has so far consumed over 1,197,694,507 tokens of AI computation on the translation alone. It is free to read, with no account required. The dated record, and the full story of how it was built, is on the blog: The First Complete AI Translation of the Bible from Hebrew and Greek.
Every English word traceable to a source token.
IIThe Committee
One model runs as eight roles that check each other.
The engine is Grok (grok-4.3, xAI). Human Bible translation is done by committees so that no single translator's judgment stands unchecked, and the APB follows the same discipline: one model, run as eight separated roles that check each other, with deterministic checks in code between them. The structure exists to catch the model copying familiar English instead of translating the source.
1
Philology
A blind analyst produces a literal reading grounded word by word in the source tokens, and flags the hard places.
2
Apparatus
Lexical and grammatical notes are assembled for the passage, with in-text cross-references.
3
Translation
Two independent candidate renderings are drafted (one close and careful, one clear and formal) each grounded in the analyst's reading.
4
Selection
A deterministic scorer, written in code, picks the lower-risk candidate and flags residue for repair.
5
Adversarial audit
The English is translated back and hunted for additions, omissions, and imported ideas. Any English word with no source token is caught and reversed.
6
Terminology
One canonical English spelling per proper name, applied corpus-wide; the tetragrammaton is rendered "the LORD".
7
Style & proofread
Readability is improved without changing meaning, then a final grammar and consistency pass runs.
8
Publish gate
A deterministic scan blocks publication until every verse passes.
IIIBlind by Design
The translator is never told it is reading the Bible.
Earlier versions of the APB showed a failure particular to translating Scripture with AI: when a model knows the text is the Bible, it imports traditional English readings instead of translating the source. The clearest example was Mark 1:1, where the model added "the Son of God" although the Greek source data used here lacks the phrase. It was reciting the Bible it already knew.
The committee's translating stages are therefore not told the text is Scripture. They receive tokens, parsed grammar, lexicon glosses, and neutral structural instructions, with no book name, no verse reference, and no mention of the Bible. The model can often infer what it is reading, and the enforcement does not depend on preventing that: the grounding contract and the adversarial audit reverse any word that has no source token, whether the model recognized the text or not.
IVLimits
AI translation has limits.
This page does not claim that AI translates better than the human committees behind the major English Bibles. The APB shows its work at a scale no human committee has published: every rendering carries the source tokens and the reasoning behind it, and the reader can check any verse.
Every rendering passes the deterministic checks and the adversarial audit before publication, and any verse can be inspected in the reader. Where the text requires a judgment call, the alternatives and the reasoning are recorded there.
VThe Record
Every decision is kept in the reader.
In the reader, any verse opens into its original text: the Hebrew or Greek, the parsing of every word, cross-references, and the translation record. The same source data feeds a lexicon of every lemma in the biblical corpus, an atlas of every place, and an audio Bible in English, Hebrew, and Greek. The Firmament (the knowledge map built over this corpus) documents its full ontology and provenance model on a specification page.
Blind retranslationThe translator sees the source tokens and nothing else
The whole text was rebuilt by a translator that never sees a reference, a book name, or an existing English version, so that no familiar phrasing could be copied rather than translated.
Revision 3The audit committee
The text was rebuilt again under a full committee: blind translation, independent fidelity audits against the source tokens, and adjudication of every disputed rendering, with the record of each decision kept per verse.
Revision 3.1Divine titles, spelling, and quotation marks
Standardized the capitalization of divine titles across all 31,166 verses: the Holy Spirit, the Father, the Son, the Lord, and related titles are now cased consistently wherever the referent is God, with the referent of every ambiguous title adjudicated from the source tokens.
Spelling was normalized to American English, and the Paraclete of John 14 to 16 is rendered "the Advocate" to match the same Greek word at 1 John 2:1. The same revision repaired quotation marks throughout: speeches that closed too early or never closed, stray marks, and short dialogue that had none, with each speech's boundaries adjudicated chapter by chapter.
Every changed verse carries its revision note in the reader, naming the editor (Claude Fable 5.1) and the adjudicator (Grok 4.3), and no verse was reworded.
Revision 3.2Textual repairs from a whole-corpus audit
Repaired the errors found by a whole-corpus audit of the text against its source tokens, with Claude Fable 5.1 (Anthropic) as editor and no model adjudication. The lost numerals of 1 Samuel 13:1 are now marked rather than read literally as "a year old."
Wrong numbers were corrected against the Hebrew in Genesis 5:7, 1 Kings 4:32, 6:1 and 7:3, Ezra 2:28 and 2:69, Nehemiah 7:70, Ezekiel 48, and Song of Songs 8:12; a dropped place name (Song 8:11) and a dropped negation (1 Chronicles 12:33) were restored; a reading imported from the Psalm 18 parallel was removed from 2 Samuel 22:29; four gloss slips and four name spellings were corrected; and John's double amen now reads "Truly, truly" throughout. Seven verses where the Hebrew numeral is intact but contradicted by a parallel passage (2 Chronicles 22:2 and 36:9 among them) keep their text and carry a textual note instead.
Every changed verse carries its revision note in the reader, with the source tokens it rests on, and the audit scripts now run as part of every release.
Revision 3.5A critical apparatus built from open sources
The earlier text translated one printed base per testament and had no way to weigh the manuscripts where they disagree. The apparatus built for this revision draws only on open sources: the United Bible Societies' Hebrew Old Testament Text Project (4,515 Hebrew problems with their alternatives), the ketiv and qere of the Open Scriptures Hebrew Bible, the STEPBible TAHOT and TAGNT tagged texts with their per-word record of eight printed Greek editions, the SBLGNT apparatus, the Center for New Testament Restoration's transcriptions of the Greek manuscripts through the fourth century, the Robinson-Pierpont Byzantine text, and Swete's edition of the Septuagint as a versional witness.
From these, 14,871 variation units were assembled and decided blind by three readers from three companies, Grok 4.3 (xAI), GPT-6 Astra (OpenAI), and Claude Fable 5.1 (Anthropic), each seeing the readings, their morphology, and the age and class of the witnesses, but never a reference, a manuscript name, or another committee's rating; the human committees and printed editions were compared only after the vote. A reading was adopted only where two of the three agreed.
On the page, 54 verses now depart from the base text and carry a note giving the reading they left, and 45 verses of disputed passages (Mark 16:9 to 20, John 7:53 to 8:11, Luke 22:43 and 44, Acts 8:37, and Romans 16:24 among them) are set in brackets with their wording kept. The same three readers decided the casing of every ambiguous title (Spirit, Father, Son, Lord, King, Rock, Shepherd) per occurrence from the source tokens, recasing 115 verses.
The 174 pericopes those decisions touch were retranslated over the decided text by the blind committee, audited for fidelity by GPT-6 Astra as a second model family with 93 minimal corrections adopted after a clean re-audit, read by eight blind English reviewers, and re-recorded for the audio Bible. Every changed verse carries its revision note in the reader, the full record of every departure with each reader's reasons is published with the release, and the apparatus tables are released CC BY-SA 4.0 in keeping with their sources.
Read the APB
All 66 books, free, no account, with parallel Hebrew and Greek and synced audio.
Biblical Lexicon
Every Hebrew and Greek word in the Bible, with frequency and every occurrence.
Biblical Atlas
Every biblical place on a real map, linked to the verses that mention it.
The Crux Registry
The places where the Bible's oldest manuscripts disagree, with every witness listed and the committee's decision at each one.
The Firmament
A knowledge map of people, places, and themes across Scripture, each link backed by cited evidence.
The Dataset
The full text (31,166 verses) as open data on Hugging Face, CC BY 4.0.
VIQuestions
Common questions
Is there an AI-translated Bible?
Yes. As far as I can determine, the Anselm Project Bible is the first complete AI translation of the entire Protestant canon, all 66 books, translated from original-language data and published in October 2025. If you know of an earlier complete one, tell me and I'll amend this page.
Is it a real translation or a paraphrase of existing English Bibles?
It is translated from per-word Hebrew and Greek lemma and morphology data. No English version is used as a source. The translating model works blind (it is not told the text is Scripture) and an adversarial audit reverses any English word that has no source token, so that traditional English renderings cannot leak in.
What is it translated from?
The Hebrew Bible text and morphology come from the Open Scriptures Hebrew Bible, based on the Westminster Leningrad Codex. The New Testament comes from the SBL Greek New Testament with MorphGNT morphological analysis.
Which AI does the translating?
Every stage runs on Grok (grok-4.3, xAI), as eight separated roles (philologist, annotator, translator, selector, auditor, terminology editor, stylist, proofreader) with deterministic checks between them, so that no single pass produces the text unchecked.
Can AI translate the Bible accurately?
It translates what the source tokens say, and the reader can verify that against the tokens shown for every verse. Where a passage requires a judgment call, the alternatives and the reasoning are recorded in the reader.
How is this different from AI-assisted Bible translation projects?
Most such efforts use AI to speed up human translation committees, usually for languages that lack a Bible. In the Anselm Project Bible, an AI committee produces the translation under deterministic checks, and readers audit it.
Is it free to read?
Yes. The full text, the audio Bible in English, Hebrew, and Greek, the lexicon, and the atlas are free, with no account required.
The text is free to read.
The full text, the audio Bible, the lexicon, and the atlas are free, with no account required.