Submit suspicious election content
Capture suspicious civic claims with source, country, language, platform, urgency, and election context.
Civic risk triage for election periods
CivicGuard Africa
AI-assisted election disinformation triage and multilingual safety benchmarking for African contexts.
CivicGuard Africa is a civic AI safety tool and evaluation framework for structured incident triage, human review workflows, civic risk monitoring, and benchmark-driven model safety analysis.
What CivicGuard does
A working tool and evaluation framework, not just a landing page
The MVP demonstrates how a civic team could move from intake to evidence review, risk monitoring, and reusable AI safety benchmarking while keeping the system local and explainable.
Generate transparent triage report
Use deterministic mock scoring to assign potential risk, harm category, confidence, checklist, and reviewer next steps.
Support human/community review
Move incidents into a reviewer queue where local knowledge, evidence, and safety judgment shape the next action.
Monitor trends and evaluate model safety
Track civic risk patterns on the dashboard and use the Benchmark Lab to evaluate multilingual AI safety behavior.
What makes CivicGuard different
Built as both a monitoring tool and an AI safety benchmark
The MVP is intentionally simple, but it connects the parts judges need to see: a working incident flow, human review, dashboard signals, and reusable evaluation data.
Full civic workflow
It combines monitoring, triage, human review, dashboarding, and benchmarking.
Africa-specific context
It is designed for African election contexts and multilingual/code-switched language.
Human-centered safety
It does not claim automatic truth detection; it supports human reviewers with structure, prioritisation, and next steps.
Reusable evidence layer
It produces reusable exports for researchers, journalists, and civic-tech teams.
Why this matters in African elections
Disinformation moves across languages faster than verification capacity can scale
AI-generated disinformation can spread faster than verification teams, especially across local languages, low-resource contexts, informal messaging channels, and high-pressure election periods. Election risks are especially sensitive in African contexts, where local languages and code-switched communication can expose weaknesses in English-centered AI safety systems. The problem is not just whether a claim is false; it is whether the right people can spot the civic risk quickly enough to respond responsibly.
Local-language pressure
Claims can shift between English, French, Swahili, isiXhosa, isiZulu, Afrikaans, and code-switched community speech.
Harm-first triage
Reviewers need to see voter suppression, intimidation, impersonation, false results, and media manipulation signals before they become larger civic risks.
How CivicGuard works
A practical workflow for civic verification teams
Step 1
Submit suspicious content
A journalist, student, civic-tech team, or community verifier records the claim and context.
Step 2
AI-assisted risk triage
The local mock engine scores risk using urgency and civic-harm keywords.
Step 3
Human evidence review
Reviewers preserve evidence, verify source context, and decide what needs escalation.
Step 4
Dashboard and benchmark insights
Teams monitor patterns and test model behavior across languages and harm categories.
Judge demo flow
A seven-step walkthrough for the final presentation
Use this sequence to show the full product loop from intake to evaluation without leaving the local MVP.
- 1Open Submit.
- 2Click Load demo election case.
- 3Generate the triage report.
- 4Review risk, category, checklist, and safety disclaimer.
- 5Open Community to see the case in the review queue.
- 6Open Dashboard to see updated metrics.
- 7Open Benchmark Lab to see the reusable AI safety evaluation layer.
CivicGuard does not determine truth automatically. It supports human verification by prioritising and structuring suspicious civic content.
Benchmark Lab