Toward Collective Action: Roundtable on Safe & Trusted AI

Key Topics Covered

• Defining "Safe & Trusted AI" for Africa – contextual understanding beyond Western safety frameworks
• Short-term vs. Long-term AI Risks – prioritizing democracy threats and election interference over speculative scenarios
• Digital Colonialism & Data Sovereignty – avoiding extraction of African data and value concentration outside the continent
• Human Agency & Cognitive Erosion – maintaining decision-making autonomy and preventing over-reliance on external AI systems
• AI Capacity Building & Access – need for African researchers to access frontier models for evaluation and benchmarking
• Policy & Governance Gaps – Africa has strategies but lacks concrete AI policies and regulatory mechanisms
• Critical Infrastructure Integration – safeguards needed before deploying AI in government, healthcare, energy systems
• Regional Collaboration – moving from competitive nation-state approaches to continental cooperation
• Public Awareness & Literacy – 75% of South African respondents have minimal AI knowledge; need for mass education
• Procurement & Accountability Mechanisms – using government purchasing power as leverage; establishing independent AI safety institutes

---

Key Points & Insights

1. African AI Risks Differ from Global Discourse – The panel reframes "existential risk" for Africa as threats to democratic integrity, social cohesion, and cultural survival rather than speculative AI control scenarios. Misinformation/disinformation campaigns leveraging AI during elections pose immediate, verifiable threats in Ghana, South Africa, Nigeria, and Kenya.

2. Digital Neocolonialism is the Foremost Threat – Ambassador Tiggo emphasizes that if AI systems extract African data, concentrate value outside the continent, and reduce African institutions to "implementers or users," this constitutes an existential (unacceptable) harm. Models built without African knowledge, wisdom, and cultures risk "civilization extinction."

3. Knowledge Gaps Prevent Meaningful Participation – Only ~25% of South African survey respondents demonstrated basic AI awareness; most learning occurs through informal/unstructured channels (social media, television). This prevents African publics from making informed decisions about AI deployment in their societies.

4. Malicious Agents Are a Democratized Threat – Individual actors can now design AI agents to conduct misinformation campaigns at scale. This is no longer limited to large tech companies—the barrier to weaponized AI deployment has collapsed.

5. AI Incident Database for Africa is Missing – Current AI incident tracking databases are biased toward Western contexts; incidents in Africa (e.g., flawed AI-based exam grading in Nigeria/South Africa) remain undocumented, making evidence-based policymaking impossible.

6. Access ≠ Capacity – African scientists cannot evaluate frontier models without direct access AND funding/infrastructure. Kenya is the largest user of ChatGPT for emotional advice, yet no Kenyan institution can audit whether those models understand Kenyan context.

7. Procurement is a Strategic Leverage Point – Governments wield purchasing power at "the point of decision" (when vendors most want to negotiate). Building safety benchmarks, audit rights, and contingency clauses into procurement documents is more achievable than post-deployment regulation.

8. "Move Fast and Break Things" is Incompatible with African Infrastructure – Premature AI deployment in critical systems (education, healthcare, energy grids) risks locking in inappropriate solutions, eroding human skills, and increasing dependency on foreign vendors who cannot be easily replaced or litigated against.

9. The Continent Must Build Its Own Models – Over-reliance on Gemini, Claude, GPT, and Anthropic models means no genuine autonomy. Building local, context-aware models (possibly on open-source foundations) is framed as fundamental to agency and sovereignty.

10. Collaboration Structures Already Exist – Grassroots initiatives (Masakhane, Deep Learning Indaba, AI4D, Sisonke Biotech, etc.) demonstrate that limited resources can produce outsized impact through network effects. Scaling these with infrastructure investment (e.g., UCT's African Compute Initiative) multiplies capability without requiring competition.

---

Notable Quotes or Statements

Ambassador Philip Tiggo (Kenya Special Envoy on Technology):

"If AI systems are creating a dependency rather than building capacity or capability, that's undesirable. The erosion of human agency, especially for a continent still trying to aspire, is a problem. If we extract African data, capture our markets, and concentrate value outside the continent while leaving our institutions as mere implementers or users, that's digital neocolonism. If models continue to be built without our knowledge, wisdoms, and cultures, it creates an existential threat—not just undesirable, but unacceptable."

Dr. Chinasa Oko (UN Office for Digital & Emerging Technologies):

"Finding basic information about AI harms on the continent is still very hard if you're not tuned in. Nigeria and South Africa had issues with AI being used to automatically grade standardized exams, and students couldn't rebut their scores. This did not make mainstream news. We need an AI incident database for Africa so governments can craft regulations that serve community needs and hold responsible parties accountable."

Professor Jonathan Shock (UC Director, UCTI Initiative):

"Misinformation and disinformation campaigns are allowing a breakdown in trust at scale. We're seeing gendered disinformation targeted at female politicians. Technology-facilitated gender-based violence is massive. The real fear is not speculative AI control scenarios—it's the collapse of democratic trust happening right now in Ghana, South Africa, Nigeria."

Ambassador Tiggo (on policy implementation):

"If we have a couple of AI strategies in the continent, we do not necessarily have AI policies. There's no mechanism. We don't have the talent. Public sector officials think AI is just ChatGPT. Safety is so far off the scale they're not even thinking about it. This needs to be an all-in effort where the dichotomy between civil society and governments disappears—because this is an existential risk to society."

Ambassador Tiggo (on continental cooperation):

"AI is not about who builds the best data centers. This is a collective, all-in effort. That's the biggest shift we need to make: from competition to cooperation and collaboration. I'm saying this out of frustration because I see it as a waste of money and resources."

---

Speakers & Organizations Mentioned

Panel Members:

• Dr. Chinasa Oko – Founder, Techruption; Policy AI Specialist, UN Office for Digital & Emerging Technologies
• Ambassador Philip Tiggo – Special Envoy on Technology for the President of Kenya
• Professor Jonathan Shock – Associate Professor, Department of Mathematics & Applied Mathematics, University of Cape Town; Director, UCTI Initiative
• Mark Gaffley – Director of Legal and Operations, Centre for Global AI Governance (GCG)
• Dr. Kola Idesen – Research Director, Research ICT Africa (noted as joining later)

Moderators:

• Zach (moderating first portion)
• Michelle Malonza (moderating Q&A portion)
• Iman (concluding remarks)

Organizations & Initiatives Referenced:

• Elina (co-hosting organization)
• AI Safety South Africa
• UN Office for Digital & Emerging Technologies
• Centre for Global AI Governance (GCG) – South Africa
• International Scientific Panel on AI (UN)
• Africa AI Council
• AU Continental AI Strategy
• University of Cape Town (announcing African Compute Initiative)
• Masakhane
• Deep Learning Indaba
• AI4D
• Sisonke Biotech
• NUMA (registration platform)
• Slider (Q&A platform)
• World Bank

---

Technical Concepts & Resources

AI Models & Systems Referenced:

• ChatGPT (OpenAI) – noted as Kenya's most-used emotional advisory tool
• Claude (Anthropic)
• Gemini (Google)
• Open-source language models (general reference)

Technical Infrastructure:

• High-Performance Computing (HPC) cluster
• GPU compute resources
• Cloud platforms
• Data centers

Research/Evaluation Frameworks:

• AI incident databases (missing for Africa; mentioned as aspirational)
• Safety benchmarks (in procurement)
• Model evaluations & audits
• Public awareness/perception surveys (e.g., South African social attitudes survey – 75% low AI awareness baseline)

Policy/Governance Tools:

• Procurement frameworks with safety clauses
• AI safety institutes (model: US NIST – National Institute of Standards and Technology)
• Watermarking for AI-generated media (discussed as insufficient solution)
• Contingency planning & multi-vendor strategies
• Media literacy & public education initiatives

Key Data Point:

• 64% of Africa lacks internet access (raising digital exclusion concerns)
• Median age on continent: 19.7 years (noted as high youth engagement with AI tools)

Methodologies & Initiatives:

• Short courses on ethical and human rights implications of AI (through accredited South African universities)
• MOOCs (Massive Open Online Courses) with relatable imagery for African contexts
• Women-focused scholarship programs in AI ethics
• Paris AI Action Summit (referenced collaborative effort)
• Global AI Dialogues (UN initiative)

---

Document Type: Conference Roundtable Transcript
Focus Area: African AI governance, safety, and policy
Duration: ~47 minutes (30 min panel + 15 min Q&A + 2 min closing)