Amanah

Trust & safety for donation crowdfunding

AI reads every campaign.
People make every call.

Amanah (أمانة — “trust”) triages incoming campaigns for fraud signals, policy violations, and zakat-eligibility claims. It scores risk, quotes its evidence, and admits what it can't verify — then hands the decision to a human reviewer. Every time.

Live demo — real AI pipeline, fictional campaigns, demo sign-in shown on the login page.

URGENT: my family will be on the street in 48 hours

Risk 91
Urgency manipulationUnverifiable claimsFinancial anomaly

“The bank has given us 48 hours before they change the locks” — cited from the campaign text

AI suggests: escalateconfidence 0.28 — low

The decision stays human

ApproveRejectEscalate

Recorded to an immutable audit log with the AI's reasoning at decision time.

Two stages of machine scrutiny. One human decision.

Stage 1 · every campaign

Screening

A fast model scores fraud risk 0–100 against a seven-category rubric — urgency manipulation, story inconsistency, financial anomalies, zakat-eligibility doubt, and more — and reports how confident it is.

Stage 2 · when warranted

Deep review

Risky or uncertain campaigns get a second pass with the organizer's history. Every concern must quote the exact campaign text or data field that triggered it — no vibes, only evidence.

Always

Human decision

Reviewers approve, reject, or escalate from a triage queue sorted by risk. The AI cannot change a campaign's status — there is no code path for it. Every action lands in an immutable audit log.

The line is drawn in code,
not in the UI.

Status transitions exist only behind reviewer-authenticated endpoints. AI assessments are served exclusively to the reviewer console — the public status page never carries a risk score, a flag, or a recommendation, because teaching submitters what the model looks for would teach bad actors how to evade it.

“Insufficient information to distinguish an urgent legitimate need from fabricated urgency… this is exactly the kind of case a human should decide.”

— an actual deep-review output from the demo queue. Calibrated uncertainty is a feature, not a failure.

The AI owns

  • Reading every submission, instantly
  • Scoring risk against a fixed rubric
  • Quoting evidence for every flag
  • Saying “I'm not sure” out loud
  • Ordering the queue so people look at the right things first

Humans own

  • Every approve, reject, and escalate
  • Judgment on genuinely ambiguous cases
  • Zakat-eligibility calls that need documentation
  • Accountability — names in the audit log
  • The rubric itself

100%

decisions made by humans

2 stages

of AI scrutiny

30 cases

labeled eval set

7

risk categories

Try both sides of the table.

Submit a campaign and watch the pipeline work in real time — then sign in as a reviewer and decide its fate. Your username shows up in the audit trail.

Amanah — AI triage copilot for campaign trust & safety