Safety in autonomous systems
Investigating how AI agents can be evaluated, monitored and constrained, with effective human intervention when behaviour becomes unsafe.
Agentic AI · Runtime assuranceAfrica Institute for Safe and Trustworthy AI
AI that earns
our trust.
Built for Africa.
Advancing research, evaluation and evidence-based assurance so AI systems work safely, fairly and reliably in African contexts.
African perspectives. Shared responsibility.
RESEARCH WITH PURPOSETechnical evaluationAgentic AI safetyEvidence-based assuranceAfrican contextual relevance
01 / The institute
Shaping AI progress.
Strengthening its safeguards.
AISTAI is an emerging initiative focused on safe and trustworthy artificial intelligence across Africa. Our ambition is to connect technical research with the evidence, institutions and human judgement needed for responsible deployment.
We place African languages, communities and operating environments at the centre of evaluation. From healthcare tools to financial decisions and autonomous agents, systems must be tested where their consequences are felt.
02 / Our focus
From questions
to testable evidence.
Our developing programme connects AI safety research, practical evaluations and assurance methods.
AI in its real context
Examining reliability, fairness and relevance across African languages, datasets and deployment settings, including high-impact uses.
LLM evaluation · Bias · ExplainabilityEvidence for better decisions
Developing practical controls and evidence guidance that help institutions assess AI, identify gaps and prioritise remediation.
AAIAF · Governance · Oversight03 / The framework
African AI
Assurance Framework.
AAIAF brings technical evaluation, governance and lifecycle monitoring into a structured assurance approach. Its eight domains connect evidence to decisions, with critical safeguards that cannot be offset by an aggregate score.
Developed through CognitiveAI Labs and being refined through expert feedback. AISTAI’s programme will build on this work through further research and evaluation.
Explore Assurance Cloud ↗D1Performance, reliability
and robustness
and robustness
D2Explainability
and transparency
and transparency
D3Fairness and
non-discrimination
non-discrimination
D4Privacy and
data governance
data governance
D5Security
and resilience
and resilience
D6Human oversight
and accountability
and accountability
D7Agentic control
and autonomy
and autonomy
D8Runtime monitoring
and incident management
and incident management
04 / Our approach
Trust is a question
of evidence.
01
Evidence before trust
Claims about an AI system should be supported by relevant, measurable and independently reviewable evidence.
02
Safeguards beyond scores
A strong average result cannot compensate for a failed critical safeguard. Human accountability and meaningful intervention matter.
03
Assurance across the lifecycle
Evaluation continues after deployment through monitoring, incident learning and reassessment as systems and contexts change.
An invitation to collaborate
Help shape safe and
trustworthy AI for Africa.
We welcome conversations with researchers, institutions, civil society and technical practitioners interested in evaluation methods, framework review and future pilot collaborations.