AI Safety South Africa (AISSA)

AI Safety South Africa (AISSA) is a Cape Town-based AI-safety capacity-building organization co-founded by Leo Hyams and Benjamin Sturgeon, operating from a co-working and events space at Innovation City under the parallel brand AI Safety Cape Town. AISSA positions itself as "a hub for AI safety work on the African continent", developing skills, networks, and community in South Africa through reading groups, structured courses, research fellowships, and hackathons. It is the corpus's first sub-Saharan African local-group entry whose central activity is insider AI-safety community-building — adjacent in form to AI Safety 東京 in Japan, and the African complement to the PauseAI national chapters the corpus already covers.

Founding and structure

AISSA grew out of an earlier Cape Town community effort — branded AI Safety Cape Town — that Leo Hyams built alongside Benjamin Sturgeon starting in 2023. By August 2024, the organization had a fast-growing community of around a hundred members and had begun formalizing its programs; the Substack newsletter published that month still used the "AI Safety Cape Town" masthead while the organization was already operating under both the AISCT events-space brand and the AISSA nonprofit brand. The two websites (aisafetysa.com and aisafetyct.com) remain live as parallel surfaces: the national nonprofit brand and the physical co-working space brand respectively.

Hyams serves as Executive Director; Sturgeon, who is pursuing an MPhil in Applied Mathematics at UCT with a focus on AI safety and agency, serves as Strategic Advisor. The Innovation City co-working space in Cape Town provides the physical anchor for community members to work alongside one another; the website notes plans to expand presence to Johannesburg and Stellenbosch. A September 2025 EA Forum post by Hyams describes AISSA as having worked alongside the University of Cape Town to integrate AI safety into the university's curriculum since the organization's inception.

Programs

AISSA's activity stack spans from accessible community entry points to supervised research:

Reading groups and community events. The organization runs regular in-person reading groups and workshops — a weekly paper discussion and a biweekly core-member reading group — alongside broader community events. The Luma calendar records 49 events through mid-2026 with 1,570 total participations.

Structured courses. AISSA offers two course-format programs. Intro to Transformative AI is a five-day intensive developed in collaboration with BlueDot Impact. Intro to Cooperative AI is a thirteen-week course that ran its first cohort of 16 participants in Q2 2025. A community volunteer program called Stellies AI Safety extends reach to the Stellenbosch university town.

Cooperative AI Research Fellowship. The flagship program for 2026 is the Cooperative AI Research Fellowship — a three-month full-time in-person research program (January–April 2026, Cape Town) run in collaboration with the Cooperative AI Foundation (CAIF), the UCT AI Initiative, and Principles of Intelligence (PrincInt), with Lambda providing GPU compute. The 2026 cohort comprised ten fellows drawn from diverse global backgrounds, mentored by researchers from Google DeepMind, Oxford, MIT, Carnegie Mellon University, and the University of Washington. The program is explicitly designed to support the establishment of UCT's African AI Safety Hub as well as AISSA's own emerging research ecosystem.

Hackathons. In March 2026 AISSA co-hosted an AI Control Hackathon with Apart Research — a three-day event focused on building control protocols, evaluation tools, and monitoring systems for AI safety, with winning teams presenting at ControlConf Berkeley in April 2026. An earlier collaboration with Apart Research produced the Global South AI Safety Hackathon, a distributed event drawing participants from Latin America, Africa, and Asia.

Research. AISSA has produced six research outputs by mid-2026, including "HumanAgencyBench" (presented at the HEAL workshop at CHI 2026) and "Precursors, Proxies, and Predictive Models for Long-Horizon Tasks" (NeurIPS 2025), alongside evaluations commissioned by the UK AI Safety Institute.

Place in the movement

AISSA is the corpus's first sub-Saharan African local-group entry and its first African entry whose primary identity is insider AI-safety capacity-building rather than civil-society digital-rights advocacy of the kind anchored by Kenyan and Nigerian organizations elsewhere in the corpus. Its founding moment — a small Cape Town reading group in 2023 — parallels the formation of AI safety community groups in other non-Western hubs (Tokyo in 2023, Nairobi-adjacent groups, Budapest) and AISSA's development trajectory from informal gathering to formal nonprofit with a research program mirrors the broader pattern of the field-building layer of the AI safety movement maturing outside traditional Anglo-American academic centers.

The organization's framing as "a hub for AI safety work on the African continent" is notable alongside UCT's September 2025 launch of the African Hub on AI Safety, Peace and Security — a formal academic initiative whose development AISSA's fellowship program was partly designed to support. Where the UCT hub operates at an institutional research and policy level, AISSA functions as the volunteer and practitioner community layer: the entry point through which mid-career professionals, students, and researchers in South Africa can begin engaging with the technical and governance AI safety literature and connect into international networks. The EA Forum post by Anthonio Oladimeji from March 2024 — "The AI Safety Conversation Is Missing 1.4 Billion People" — named the structural gap AISSA is positioned to address: weak African representation in global AI safety governance, value alignment, and regulatory capacity. AISSA's funders include Open Philanthropy, whose backing of the organization situates it within the EA-adjacent AI safety philanthropic ecosystem.