Watch the interview
Control is written into the deal
In the supplied interview material, Gebru argues that African countries can give technology partners broad access to data before establishing how local institutions and communities will benefit. Data collected for a public service can also become an input for products governed elsewhere if the agreement permits secondary use.
Right Livelihood named Gebru a 2026 laureate for work challenging concentrated power in AI. The African Union’s Continental AI Strategy, endorsed in Accra in 2024, calls for African-led development and identifies computing capacity, datasets and skills as priorities. Sources and attribution
Hosting, training rights and operational control can sit with different parties. A facility can be physically local while its operator retains control over access, model updates or migration. Cross-border partnerships can preserve meaningful local authority when agreements make it enforceable.
Local value depends on whether the project funds local engineering and research, gives local institutions a say in procurement and deployment, and accounts for the labor, electricity and water the project uses.
Where control can shift
- CollectDefine the purpose and the data that may be collected.
- ProcessIdentify who can access records and where processing occurs.
- TrainSpecify whether records can be reused to develop models or products.
- Deploy and exitAssign authority over updates, audits and migration.
Five safeguards, three scenarios
What each example includes
- Vendor-managed pilot includes purpose limits.
- Governed cloud partnership adds approval for model training and independent audit access.
- Partnership with all five safeguards also includes tested portability and a funded local skills plan.
What to look for in the agreement
| Safeguard | Question | Evidence to request |
|---|---|---|
| Purpose limits | Can records be reused, transferred or shared with subcontractors? | Permitted-use schedule, processor list and retention terms. |
| Training approval | Who approves use of local data for model training or fine-tuning? | Written approval rules covering vendors, affiliates and downstream use. |
| Independent audit | Who can check how data was used and investigate harms? | Audit rights, access to relevant logs and a complaint process. |
| Tested portability | Can the institution move its data and keep the service running? | Export formats, migration test, exit costs and deletion verification. |
| Funded local skills | Who will operate and maintain the service after handover? | Training budget, handover milestones and named local roles. |
Sources
- Interview clip shared by @journogeorja · The supplied screenshot identifies the interview and provides the publisher’s summary.
- Right Livelihood laureate profile · Award and biography.
- African Union Continental AI Strategy · Continental policy context.
The contract scenarios and procurement table are editorial analysis.


Leave a Reply