Twelve African Labs Sign a Shared AI Safety Charter — Enforcement Is the Open Question

Twelve research labs from seven countries signed the Continental AI Safety Charter in Accra on Friday, committing to pre-deployment capability evaluations, a 72-hour incident disclosure window, and third-party red-teaming for any model released for public use.

The signatories include most of the continent's better-known labs — Kalahari, Msingi, Okavango, and Serengeti Analytics among them — and together account for a large share of the models actually shipping in African languages. Notably absent are the multinational providers whose models most African developers still build on.

"A charter that binds the twelve labs least likely to cause a serious incident and none of the ones that might is a start, not a finish," said Nkechi Balogun of the Abuja-based Center for Digital Rights, who reviewed the draft. "The signatories know that. Several of them told me so."

Enforcement is the acknowledged hole. There is no secretariat, no audit function, and no consequence beyond removal from the signatory list. Supporters argue that establishing a shared vocabulary now makes future regulation easier to draft, since lawmakers can point to a standard the industry already claims to follow. Skeptics counter that voluntary codes have a habit of becoming a reason not to legislate at all.