Forever, O Lord, your word is firmly fixed in the heavens.
~ Psalm 119:89 (English Standard Version)
AI has changed the stewardship of God’s word forever. Ancient tradition is merging with cutting edge technologies. As more people — especially in the Western world — turn first to a chatbot to learn about the world around them, the Church has an urgent mandate to ensure that they find Scripture there. This is a critical juncture for both believers and seekers alike.
Now is the time for action. Scripture is a key resource that will be used by AI systems, and only with your proactive efforts can it be done in a way that honors the text and preserves it for generations that will soon only know systems that depend on AI. The attached study has empirically proven beyond a shadow of doubt that faithful quotation of Scripture is possible in AI systems. We urge you to loosen your overly restrictive posture toward the use of your Bible translations in generative AI contexts. The Apologist AI platform has been successfully utilizing the techniques described in this paper for the past 6 months with public domain and less restrictive Bible translations.
While it is true that some conversational AI applications may be created without the necessary safeguards in place, this is not proper cause to deny licenses to faithfully stewarded technology platforms that incorporate generative AI with discernment. The abusive and careless use of God’s word is not a possibility unique to this generation; it is an inevitability in every generation ever since the days of Adam.
In spite of the inherent risks, there is also great reward. We invite you to help us shine the light of God’s word in the digital spaces where it’s currently lacking. The shift from traditional search to AI-centric content discovery is happening quickly, and time is of the essence if we are to continue to meet God’s image-bearers where they are.
FAQ
Why do Bible translation licenses currently prohibit generative AI? Because AI is probabilistic, and licensors have reasonably treated that variability as incompatible with guaranteeing exact, faithful wording.
Is there a way to use a licensed translation without misquotation risk? Yes: output transform, where the model names only the reference and a deterministic process inserts the licensed text, tested at 100% exact accuracy in English.
Has this run in a real product? Yes, six months in production with public domain and less-restrictive translations, plus the formal study.
What is this letter asking licensors to do? Reconsider blanket AI prohibitions given a specific architecture that removes the misquotation risk, and consider narrower, method-specific terms instead of a categorical ban.
Does this ask licensors to trust AI broadly? No, it asks them to evaluate one specific, testable, open-sourced method against the actual risk their license protects.
Conclusion
The caution behind current licensing was, and largely still is, well founded. What’s changed is that the specific risk now has a measured, open-sourced fix. Read the full study, review the test harness yourself, and sign the petition.