Chapter 13: What to Refuse—and What We Are Free to Do

Refusal is not Luddism. It is love with lines.

There are uses of AI that trade away the human thickness of this work: accountable presence, truth told in relationship, prayer as address to God, formation that costs time. There are other uses—boring, holy uses—that free leaders to show up where only a body can.

The boundary is vocational, not a superstition about electrons.

This chapter names lanes; chapter seven named the seasons of an organization’s learning curve. Some lines below are permanent—never simulate intimacy that belongs to face-to-face care or ship words nobody owns. Others work as temporary fences during structured experimentation: many teams pause publishing AI-affected finished voice until disclosure and review norms are real, and pause feeding private counseling-related, HR, or donor payloads into tools their policy and IT stack have not cleared. Permission to learn does not mean permission to exploit surprise. When a fence comes down, it should be because you documented what you learned—not because the calendar bullied you into looking “current.”

Refusals (starting list, not closed canon)

1. Publishing without a responsible human
Fully automated pipelines that ship words nobody has weighed are not “efficiency.” They are moral outsourcing.

2. Simulating care you are not actually present to carry
If a reasonable reader thinks they are in relationship with you and they are mostly in relationship with a model, you have crossed from amplification into impersonation. For movement leaders, approval of AI-affected public voice belongs with the leader and whatever human circle they have deputized—not with the default settings of a vendor.

3. Synthetic intimacy
Care, confession, and direction belong to covenantal presence. A chat window can support reflection; it cannot carry the full weight of a person alone.

4. Theology without verification
Models hallucinate citations and flatten nuance. Traditioned humans must verify claims about God, Scripture, and the lives of neighbors—the almost-right failure (chapter eight) hits doctrine with unusual cruelty.

5. Deception
Hiding material AI involvement where a fair reader would care violates integrity—the same integrity chapter ten traced when Americans punish post-hoc discovery of AI in political speech, news, and faith-shaped advice. The homework there is not “never touch a model.” It is do not make people guess what passed through one.

6. Ungoverned data in the wrong container
Caseload notes, counseling summaries, personnel files, and donor histories are not generic “prompt material.” If your maturity season still has no clear boundary for what may leave your systems, the refusal is blunt: do not treat consumer tools like a private vault until governance says otherwise.

Theological integrity (non-negotiable)

For movement leaders, God-talk is never decorative. When a model smooths a hard edge of Scripture, elides tensions your tradition has learned to hold honestly, or invents a citation that sounds learned, the failure is not only factual—it is spiritual. People learn what God is like from the sentences you let through.

That does not mean “never draft with assistance.” It means claims about God, Scripture, and the lives of neighbors stay under human judgment tethered to whatever rule of faith your community actually runs—elders or board, supervisors, theological peers—not under whatever completion the model favors this week.

When integrity slips, you rarely get a headline scandal first. You get the almost-right paragraph (see chapter eight): fluent enough to teach, wrong enough to misshape hope. The repair is not only “better prompts” but accountable circles willing to delete sentences that scan clean and still miss the mark.

Freedoms (and some obligations)

You are free—often obliged—to use assistance for:

• Scheduling, calendaring, first-pass transcription
• Draft outlines you will rewrite
• Literature sweeps you will verify
• Translation support reviewed by bilingual leaders
• Accessibility passes (alt text, readability) checked by humans
• Documented internal experiments—breaking prompts, stress-testing drafts in private sandboxes, logging “near misses”—so your team learns before those lessons become a public correction letter

Within the lanes above, you are also free—indeed wise—to amplify work you have already done: make corpus and teaching findable, invite networked correction (the scenius instinct from chapter nine), and disclose assistance where a fair reader would care (chapter ten)—so freedom does not smuggle in secrecy.

These are not “less spiritual.” They are stewardship—if they buy an hour with a teenager in crisis, buy the hour.

Denominational weather (landscape, not unanimity)

Public Christianity is arguing in real time, not issuing a single FAQ. Useful orientation points for staff conversations—not consensus proofs—include bodies that have published AI-ethics language you can compare against your own polity (always pull current text before citing them in public):

• Southern Baptist Convention and related agencies—resolutions and guidance in recent cycles urging caution, disclosure, and human accountability with generative tools.
• World Council of Churches—ecumenical process and statements on AI ethics (early-2020s onward).
• Rome Call for AI Ethics (2020; Pontifical Academy for Life and partners)—a compact human-dignity frame many Catholic and secular institutions still reference.

Use them as conversation starters, not cudgels. Your bishop, presbytery, board, or elders remain the decisive voice for what your people are bound to hear as faithful.

Nonprofits: money words need more fear of God, not less

Grant narratives, impact metrics, and donor letters are high-trust genres. AI may help outline—but numbers and claims pass through finance and program officers. The same private-data discipline that protects a counseling or case-management note protects a beneficiary story pulled from a case file: if it identifies a vulnerable person, it is not disposable training fodder. Refuse “pretty false.” Freely automate the meeting that used to steal the evening you needed to sleep.

Refusal as formation

When a community practices public refusals—calmly, with reasons—it teaches discernment more than a policy PDF ever will. When it practices public freedoms—also with reasons—it teaches stewardship without shame. The capstone of the maturity picture in chapter seven is still rare: leaders who demonstrate mature use across an organization others can imitate. Refusals and freedoms named out loud are how you move from slides about ethics to embodied credibility.

Next we go deeper into what the tool displaces when we forget those lines—and why embodied leadership remains non-negotiable.

---

Reflection questions

1. Which refusal would cost you popularity—and are you willing to pay it?

2. Which freedom would scandalize someone you respect—and how will you explain it?

3. What is your local “red line” that is not on my list—but must be on yours?