Back to Home
Research

Open research for verifiable safety.

All research is published openly. Every finding feeds directly into safety benchmarks and policy recommendations for Zambia.

ArchitectureIn Progress

Hexcore World Model Architecture

A novel architecture exploring how world models can be structured for safer, more interpretable AI systems with improved alignment properties.

World ModelsAlignmentArchitecture
Red TeamingOngoing

Frontier Model Safety Evaluation

Direct contributions to safety evaluations of GPT-4o, Deep Research, and Claude models identifying critical vulnerabilities before public deployment.

OpenAIAnthropicVulnerability Discovery
Threat IntelQ1 2025

Zambia AI Threat Landscape Report

Comprehensive mapping of AI-specific risks in Zambia: deepfake political manipulation, AI-powered mobile money fraud, algorithmic bias in financial services.

DeepfakesFraudElections
GovernanceDraft

AI Safety Policy Framework for Zambia

Practical governance recommendations aligned with the National AI Strategy (2024-2026) and African Union frameworks. Bridging global best practices with local realities.

PolicyAU FrameworkNational Strategy

All publications are open access. View our research on Zenodo.

View on Zenodo