A hub for AI safety on the African continent

I am the Founder and Executive Director of AI Safety South Africa, a capacity-building and research organisation focused on global catastrophic risk from advanced AI.

An AI safety researcher based in Cape Town South Africa. My work revolves around researching ways to reduce catastrophic risks from transformative AI systems. Currently my work focuses on understanding conceptions of agency and trying to develop mathematical models of agency so that we can better steer AI systems towards behaviours which enhance human agency rather than deplete it. I am also excited about evals that aim to address this problem.

As a graduate of the Michaelis School of Fine Art, I think deeply about visual communication and interactive experiences. In 2025 I Completed a Cooperative AI course which stoked my curiosity about the intersection of interaction design and multi-agent systems. With prior experience in User Experience Design, Art Direction and Project Management, I'm keen to leverage these skills to build the community and brand at AISSA. My aim is to design for cooperative outcomes that support human agency within AI safety.

Charl Botha is the Systems Lead for AI Safety South Africa. He holds a BSc in Computer Science and Applied Statistics from UCT. He is currently working on a research project for multi-agent simulations and evaluations, with a focus on incentivising cooperation in multi-agent settings. Charl has prior experience as a Fullstack Software Developer and has been frequently attending meetups, discussions and training in the AI Safety Cape Town Community. He is also one of the course facilitators at AI Safety South Africa.

Claude is an Affiliate Research Scientist at AI Safety South Africa, where his research focuses on Multi-Agent AI Safety with frontier models. He holds a PhD from the University of Cape Town (UCT) under the supervision of Associate Professor Jonathan Shock, a journey supported by his role as a Cooperative AI Foundation PhD Fellow. His doctoral research specialised in Offline Multi-Agent Reinforcement Learning, specifically investigating how agents can learn to cooperate and coordinate using only static datasets. A frequent contributor to the research community, Claude has published several times at NeurIPS. His academic background is bolstered by five years of industry experience as a Machine Learning Research Engineer at InstaDeep. Claude is dedicated to ensuring that the next generation of advanced AI systems is designed for robust cooperation and safety.

Jaco du Toit is a machine learning engineer and AI safety researcher based in Cape Town. After completing his master's degree and spending a year in industry, he pivoted to focus on AI capability evaluations. He has developed several evals for the UK AI Security Institute through their bounty program and is currently working on evals aimed at autonomous replication and adaptation risks.