FideAI

FID-024 · Open question

Faith-AI Incident Database

Should Fide AI maintain a public or semi-public incident database for failures, near misses, misuse, and disputed deployments of AI in faith-facing contexts?

Why the question remains open

Safety fields learn from incidents. Faith communities are already encountering AI-generated sermons, spiritual chatbots, pastoral automation, synthetic clergy, and religious misinformation. Without a shared incident vocabulary, institutions will repeat mistakes.

Working hypothesis

A proposition to test, not a finding.

A carefully governed incident database can reveal recurring patterns and improve standards, but only if it protects privacy, avoids sensationalism, distinguishes verified from reported incidents, and gives organizations a correction process.

Proposed method

How the question could be tested

  • 01Define incident categories and severity levels.
  • 02Create intake, verification, redaction, and publication workflows.
  • 03Track source type, affected population, system type, failure mode, evidence quality, and remediation.
  • 04Publish aggregate trend reports before publishing sensitive individual cases.

Needed controls

What must constrain the study

  • 01Defamation and privacy review.
  • 02Evidence-quality labels.
  • 03Right-of-reply and correction process.
  • 04Redaction of vulnerable-user details.

Expected outputs

Artifacts the work should produce

  • 01Incident taxonomy.
  • 02Intake form.
  • 03Public aggregate report template.
  • 04Governance policy for incident publication.

Open questions

Uncertainties the protocol must resolve

  • 01Should the database be public, private, or tiered?
  • 02What threshold makes a report publishable?
  • 03How should anonymous reports be handled?

Open question

Open work

Primary need: taxonomy design, reporting governance

  • Design taxonomy and severity levels.
  • Review legal/privacy risks.
  • Contribute incident-response experience.
  • Build intake and redaction tooling.