About Rumour Shield
Verification infrastructure for a messaging-first society, where rumours move faster than institutions can respond.
No traceable evidence, no confident verdict.
Our Mission
Misinformation spreads fast — especially during elections, health crises, and economic uncertainty. Rumour Shield lets anyone verify a claim in seconds, in Hausa, Yoruba, Nigerian Pidgin, English, Arabic, or a mix — over the web, WhatsApp, or Telegram.
We believe an informed citizenry is the foundation of a healthy democracy.
Evidence-first, not memory
The goal isn’t an AI that claims to know the truth. It’s verification infrastructure: multi-agent evidence retrieval, narrator-chain credibility grading, a cautious policy gate, and human review for what’s left uncertain.
Privacy-first
Obvious private data — phone numbers, emails, tokens — is redacted before verification ever runs. Public reports and WhatsApp/Telegram sender references are hashed, never stored or shown in the clear.
Auditable
Every verdict is backed by a signed, hash-chained trust ledger and a shareable QR receipt — verify the chain yourself, independent of us.
How a claim moves through the system
- 1A claim arrives via the web form, WhatsApp, or Telegram — as text, a link, or a media attachment.
- 2Obvious private data (phone numbers, emails, tokens) is redacted before anything else happens.
- 3Language is detected (Hausa, Yoruba, Nigerian Pidgin, English, Arabic, or a mix) and local-language terms are expanded for search. Any shared link gets SSRF-safe structural risk analysis — shortener, typosquat, and official-domain-impersonation detection.
- 4The claim is classified by domain and reformulated into a precise, checkable statement.
- 5Evidence is retrieved from trusted sources, a human-curated evidence library, and — where configured — live web search, then deduplicated and scored for credibility.
- 6Each piece of evidence is traced back to its origin and the resulting chain is graded by its weakest link, isnād-style.
- 7A policy gate stays cautious: with no traceable evidence, or a time-sensitive claim without current-enough evidence, the verdict is downgraded rather than guessed.
- 8High-risk or low-confidence cases are queued for a human reviewer.
- 9A signed, hash-chained verification receipt is issued, with a shareable QR code and a public report.
Channels
The same pipeline serves the web app, WhatsApp, and Telegram. Sender identity and message references are hashed, never stored in the clear.
Human review
Claims with weak or conflicting evidence are queued for a reviewer, who can resolve or reject them with notes. Evidence strength gates review — not which topic a claim happens to be about.
Responsible Use
This tool is an aid — not a final authority. Always read the source links on the public report yourself. AI can make mistakes. Use your judgement.
A prototype under active development
Live web search, additional LLM providers, page-snapshot fetching for shared links, and wider media forensics are wired in as boundaries the system is built around — not all active by default until deployed behind the right credentials and infrastructure. Results are intentionally cautious.