Why faith contexts raise the bar
Messages inside a chancery, religious order, school leadership team, parish office, or charity may reference vulnerable people, vocational matters, donor relationships, local conflict, or unpublished decisions. Staff may underestimate risk because the content “is only a draft.” Drafts still travel. Logs still exist. Screenshots still happen. Model providers may still retain prompts under default settings.
Privacy here is not only GDPR compliance. It is also reputation, safety, and the moral duty not to expose people who trusted the institution. Catholic workplaces have long practised forms of professional discretion; AI does not cancel that tradition — it stress-tests it.
What “internal communications” includes in practice
Think beyond formal letters. Internal communications that often end up in AI prompts include:
- Email threads forwarded into a chat box for “summary”
- Meeting notes pasted for action-item extraction
- Draft replies to upset parents, donors, or parishioners
- HR or volunteer coordination messages with names and schedules
- Partial case details “anonymised” poorly (unique roles and dates still identify people)
Each of these can embed personal data or context that should never leave approved systems without a documented path.
Practical controls for internal AI use
- Written acceptable-use rules: what may and may not be pasted into AI tools, with local examples
- Separate workspaces or accounts for public marketing versus internal operations
- Prefer systems that keep conversations under organisational tenancy with revocation and admin oversight
- Disable unnecessary retention for exploratory chats; set retention deliberately for production use
- Train people with concrete bad/good examples (bad: full pastoral email threads; better: de-identified templates and role-play without real names)
- Channel high-sensitivity work back to human procedures and approved case systems
Controls fail if they are only posted once. Build them into onboarding, annual reminders, and manager expectations.
Communications you should keep off general-purpose chatbots
As a rule of thumb, keep the following out of consumer AI products unless counsel and security have approved a hardened path: safeguarding files, medical or counselling content, HR disciplinary cases, unpublished financials, authentication secrets, legal or tribunal materials, and anything subject to professional secrecy.
“Anonymise first” is not a free pass. Small communities make re-identification easy. When you need AI over institutional knowledge, use a designed knowledge base with access control — see secure internal knowledge bases — rather than pasting document dumps into a public chat.
Culture: forming staff, not only blocking websites
Technical blocks help, but culture decides whether people work around them. Explain that discretion protects persons, not bureaucracy. Celebrate careful use. Make approved tools easy enough that the secure path is the path of least resistance. That combination — usable approved AI plus clear red lines — is how faith-based organisations adopt technology without hollowing out trust.
Pastoral sensitivity without over-classifying everything
Not every email in a Catholic workplace is high risk. Over-classification leads people to ignore policy. Under-classification leads to harm. Use the simple public / internal / personal / high-confidentiality ladder, teach it with local examples, and revisit edge cases in team meetings.
When in doubt about a real person’s situation, keep the content in human channels and approved case systems. AI is optional. Dignity is not. Leaders should model the behaviour: if managers paste sensitive threads into consumer tools, policy documents will not save the culture.
Third parties and contractors
Consultants, translators, and freelancers may use their own AI tools on your drafts. Contract for confidentiality and approved tooling. A perfect internal policy fails if contractors paste your text into consumer models overnight. Provide them access to the same approved environment you use, or accept that you have expanded your processing without documentation.
Communications planning with AI assistance
AI can help draft newsletters, event notices, and routine replies when the facts are public or approved. Keep a human editor who knows the community. Automated tone that sounds corporate can still leak unfinished decisions if the prompt includes internal debate. Separate “help me write publicly” from “help me think through a confidential case.”
Examples of safer prompt hygiene
Instead of pasting a full thread with names and medical details, ask the model to improve structure using a redacted template. Instead of uploading a spreadsheet of donors, ask for general best practices for thank-you letters and apply them yourself. Instead of summarising a safeguarding file, keep that work in the approved case system with trained humans.
Small habits compound. Publish a one-page “prompt hygiene” card next to your acceptable-use policy so staff have something practical at hand.