madeenan
Writing

July 2026

Islamic AI safety checklist

A careful launch asks a simple question: what happens when the answer should not be confident?

Islamic AI features need a different launch checklist from ordinary support bots. The product may touch worship, grief, family, money, education, or personal religious practice. A fluent answer is not enough.

This checklist is for product teams deciding whether an Islamic AI feature is ready for users. It is not a religious approval process. It is a practical way to reduce avoidable product risk.

Role is clearThe product does not imply it issues rulings or replaces qualified scholarship.
Citations are visibleSources appear near the answer, not hidden behind a distant menu.
Search-only existsThe team can show sources without generated prose when that is safer.
Sensitive topics slow downThe flow changes when questions need human judgement or care.
Uncertainty is allowedThe answer can say the sources are not enough.
Review is possibleInternal teams can inspect source trails after an answer is shown.
RiskProduct responseBetter user message
Answer sounds like a rulingClarify the product role and route sensitive cases."Review these sources and ask a qualified person."
Citation is treated as proofShow the source text, not only a label."Open the cited source."
Sources are not enoughAllow refusal or search-only results."The available sources do not settle this."

1. The product explains what it is

Users should understand whether they are using search, cited explanation, drafting support, or a private research tool. Do not let the interface imply that the product gives rulings or replaces qualified scholarship.

2. Citations are visible near the answer

A source trail should not be hidden in a menu or buried at the bottom of the page. If the answer depends on a source, the user should be able to inspect that source without losing the thread.

3. Search-only mode exists

Not every question needs generated prose. Some contexts are better served by showing source material, teacher notes, or a handoff to qualified people. A safe product makes that path available.

4. Sensitive topics have a different path

Questions involving worship practice, family disputes, finance, grief, medical decisions, or legal consequences should not be treated like ordinary trivia. The product should slow down and make human guidance clear.

5. The answer can say "not enough"

A good Islamic AI feature should be allowed to admit that the available sources do not settle the question. If every response feels complete, the product is probably hiding uncertainty.

6. Reviewers can inspect the source trail

Internal teams should be able to see what sources were used and how they appeared to the user. This matters for support, quality review, and improving the product over time.

7. Attribution describes the real role

"Source-backed by Madeenan" is different from "verified by Madeenan." The first describes retrieval and citation support. The second implies religious approval, which is not the product's role.

The goal is not to make Islamic AI sound safer than it is. The goal is to design the product so users can see its limits.