EkklisiaAI for the ΑΩ

Three things your church AI policy probably doesn't cover yet

16 November 2026 · 5 min read

The v2.0 policy template covers three things most church AI policies miss — and each one takes ten minutes to fix.

Twelve days ago we published a copy-ready policy template, “A church AI policy in a weekend”: a short list of approved uses, a never-enter list for private and pastoral information, a giving-data rule, and a review date. It is still a good place to start — most churches still have no policy at all.

But the tools moved while the template sat still. On 10 November, Church Law & Tax and Gloo released Version 2.0 of their Artificial Intelligence Use in the Church Workplace Policy Template, and its list of what's new reads like a list of things a policy written even a few months ago would miss. Three of them matter for almost every church, whether or not you ever buy that template.

1. AI is no longer only the chatbot you opened on purpose

Version 1 of most policies — ours included — assumes someone chooses an AI tool and opens it. Version 2.0 covers embedded AI features and agentic AI as well: AI built into software the church already uses, and newer tools that can connect to systems, access information, automate workflows, and take actions with limited human intervention.

That changes the approved-tools list from a one-off decision into something to keep checking. A tool you approved last year may have gained AI features since, and an automation you switch on today may do more than draft — it may act.

The practical version: once, list what your existing software now does with AI — your email, your calendar, your church management system, your meeting tools. And before anyone turns on an agent or an automated workflow, one named person approves it and checks what it actually did. The template's rule still holds — a person always makes the final decision — but for agentic tools, that person also needs to see what the tool did on its own.

2. Transcription is now a pastoral decision, not a convenience

Version 2.0 adds safeguards for AI recording, transcription, summarisation, and note-taking involving pastoral care, counselling, personnel matters, children and youth, legal matters, and other sensitive communications. It also introduces “Protected Data Terms” for AI tools that may handle sensitive or confidential church information, and a more thorough risk review of providers — data retention, subprocessors, integrations, access controls, security safeguards, and incident notification.

Our template already says never to put private or pastoral information into a general chatbot. The new point is narrower and more everyday: the notetaker that joins a meeting uninvited. A transcription tool feels harmless because it only writes down what was said — but a pastoral conversation transcribed, summarised, and stored by a third-party service is pastoral information in a system nobody vetted.

The practical version: decide before the meeting, not during it. Sensitive conversations — pastoral care, counselling, a personnel matter, anything involving children — are not recorded or transcribed by AI by default. If a meeting is transcribed, say so at the start, and keep human notes for the sensitive parts. The same “never-enter” list from the weekend template now needs a second line: not just what never goes into a chatbot, but what never gets recorded by one.

3. A familiar voice asking for money is no longer proof of anything

Version 2.0 addresses AI-generated impersonation, deepfakes, and fraud, including independent verification of unusual requests involving payments, credentials, data transfers, or account changes.

This is the one to act on even if you change nothing else. A message that sounds like your pastor, in their voice, asking the treasurer to move money urgently is exactly the kind of request a church is built to honour quickly — and exactly the kind AI can now fake.

The practical version: agree one rule and tell everyone who handles money: any unusual request involving a payment, a password, or access — however familiar it sounds, whatever channel it arrives on — gets checked on a second, separate channel before anything moves. Call the person back on a number you already have. That single habit is worth more than any clause in the policy.

What to do this week — thirty minutes, not another weekend

You do not need to rewrite the policy. Three additions, agreed in one short meeting:

  1. Embedded and agentic AI: list the AI features already switched on in software you use, and require approval before any agent or automation is turned on.
  2. Transcription: no AI recording or transcription of pastoral, counselling, personnel, or children's conversations by default — and say so before a meeting is recorded.
  3. Verification: unusual payment, credential, or access requests are verified on a second channel, every time.

Then bring the review date forward. The template says review at least once a year, and whenever a tool changes. Turning on an agent counts as a tool changing.

An honest limit

This piece is built from Church Leaders' summary of Version 2.0 (10 November 2026), not from a read of the full template text, and the three additions above are borrowable principles, not legal advice. Church Law & Tax and Gloo's template is their product; Ekklisia's weekend template stays free and copy-ready, and these three points can simply be added to it. The point is not that every church needs a longer policy — it is that the shortest policy still has to cover the tools as they are now, not as they were when it was written.

Sources: Church Leaders / Church Law & Tax, “How To Update AI Policy To Address New Risks From Agentic AI, Deepfakes, and Data Security,” 10 November 2026 — Version 2.0 of the Artificial Intelligence Use in the Church Workplace Policy Template (Church Law & Tax and Gloo) expands coverage to embedded AI features and agentic AI, adds “Protected Data Terms” and a more thorough provider risk review, adds safeguards for AI recording, transcription, summarisation and note-taking involving pastoral care, counselling, personnel matters, children and youth, legal matters and other sensitive communications, and addresses AI-generated impersonation, deepfakes and fraud, including independent verification of unusual requests involving payments, credentials, data transfers or account changes. One limit, stated in the article: this piece is built from the Church Leaders summary of Version 2.0, not from a read of the full template text, so it claims nothing about the template's wording, price or availability beyond what the summary states.

Related reading: A church AI policy in a weekend — the copy-ready template this piece builds on; Austria just wrote the first binding national church AI rules — and the first rule is about people, not technology and Five traditions, one voice: the Swiss churches just published the AI guidance the others are still writing — the two policy frameworks this piece sits alongside; Responsible AI — the site's guide to privacy, accuracy and trust. This guidance is general and educational, not legal advice; your church's own safeguarding, data-protection, and denominational policies always take precedence.