Who is responsible for church AI?

The useful takeaways

1. Choose training around the work people need to do

   The programs we examined include a Portland ministry build lab, a Catholic media ethics roundtable and student discussions of church teaching. Their published plans emphasize different abilities: making a workflow, judging a communication decision and understanding a shared text. The useful connection is that a church may need all three kinds of learning, with a clear purpose for each.

   Try this: Before booking training, ask what participants should be able to demonstrate afterward. Include one check of judgment, such as explaining when to seek another person’s review, alongside any technical exercise.

   Keep in mind: These are organizers’ stated aims. Our report does not establish attendance or learning outcomes.

   Read our report: Church AI programs pair practical skills with different tests of responsibility

2. Give every boundary a responsible person

   An October 1 paper proposes a supervised religious-learning assistant and excludes sacramental confession, clinical counselling and autonomous pastoral care. It also questions names that imply clerical authority. For a church evaluating an assistant, the next question is practical: when it reaches its limit, is there a named person who can receive the question, correct an answer or pause the service?

   Try this: Write down what the proposed tool may answer, what it must refer and who receives those referrals. Use fictional questions to examine the boundary before relying on it with real people.

   Keep in mind: The paper presents design proposals awaiting empirical testing; it does not demonstrate a deployed service’s safety.

   Read our report: New pastoral-AI paper draws boundaries around religious-learning chatbots

3. Check what an official document actually authorizes

   Our comparison of Protestant church documents shows why an official link is only a starting point. PC(USA) has adopted social witness policy on AI. The ELCA’s foundational paper explicitly introduces no new social teaching and is neither final nor definitive. Those differences belong in any explanation generated from them. A summary that preserves the words but loses the document’s standing can give a reader the wrong impression of what their church has decided.

   Try this: For each source used in a church answer, keep the issuing body, document type, date and stated limits visible. Ask whether the answer clearly separates adopted guidance from discussion and local interpretation.

   Keep in mind: This comparison concerns the specific documents linked in our analysis. It does not establish a position shared by all Protestant churches.

   Read our analysis: When AI explains a church’s teaching, the document’s status matters

4. Let listeners help judge the translation

   PRAY.COM’s account of using HeyGen describes the possibility of carrying ministry video into other languages. Our September 30 analysis treats that account as a company claim; it does not test a translated sermon. Its Christian application draws on Paul’s concern for understandable speech in gathered worship in 1 Corinthians 14. For a ministry, that directs attention toward what listeners can understand and question. A familiar voice alone cannot establish that the translated teaching is faithful.

   Try this: For a proposed translation, identify who can review the receiving language and explain a correction in it. Check the speaker’s authorization separately from the accuracy of the finished version.

   Keep in mind: The product account is a company claim. Our analysis does not test a translated sermon or establish its accuracy.

   Read our analysis: When a sermon travels in a borrowed voice

   1 Corinthians 14 (NIV)