African states need to establish who bears responsibility when AI systems damage elections, not merely whether to use them, according to Professor Henry Kwasi Prempeh of the Ghana Center for Democratic Development.
“The question is not whether we embrace these technologies but under what safeguards and who is accountable when they cause harm,” he said at the opening of the Third Electoral Integrity Summit in Accra.
That framing sets a higher bar than most. Warnings about deepfakes are now routine; assigning liability for them is not.
The cases already on the record
The summit’s concerns are not hypothetical, and three instances have surfaced in the past month alone.
Anthropic’s September threat intelligence report identified a Kenyan actor using its Claude models to mass-produce political content ahead of the 2027 general election, generating batches of 50 posts at a time explicitly instructed to appear organic. The content praised a cabinet secretary over a tariff decision and pushed claims that the opposition coalition was fracturing.
Nigeria’s Defence Headquarters disowned an AI-generated video in August that falsely showed Chief of Defence Staff General Olufemi Oluyede making partisan political statements, warning that it risked portraying the armed forces as taking a political position.
Zambian police warned in June of AI-generated misinformation targeting officials ahead of that country’s elections.
Kenya and Nigeria both go to the polls in 2027.
Three problems, one mechanism
Prempeh’s central argument is that money, technology and civic space cannot be addressed separately.
On finance, he said cost estimates for African political campaigns have become outdated, with political actors indicating actual expenditure runs significantly higher than research captures. “Money in politics has become a really big problem, not just in Ghana but across the continent.”
On civic space, he warned that restrictions on journalists, election observers and civil society prevent abuses being exposed. “Africa’s electoral integrity is only as strong then as the civic space in which it is contested.”
The connection is the useful part. Undisclosed campaign money can purchase AI-generated content at scale; a constrained civic space means no one is positioned to document it. Weakness in one area reinforces the others.
Bloomwit Africa’s research on Nigeria’s 2027 vote, covered by iAfrica in July, described the same asymmetry in commercial terms: the cost of fabricating and distributing a damaging association has collapsed, while the cost of detecting and correcting it has not fallen at all.
What was called for
Prempeh’s recommendations covered stronger political finance transparency and regulation, safeguards for AI and information integrity, and improved protection and accreditation for media practitioners and citizen observers. He described defending electoral integrity as a shared African responsibility requiring cooperation among states, civil society and democratic institutions.
Reverend James Lahai, chairperson of the West Africa Election Observers Network, cautioned against treating technology as sufficient. “Political trust cannot be built on technology alone.” He called for stronger partnerships between electoral management bodies and citizen observers, and raised concern about shrinking space for election observation in some African countries.
Samuel Dottoh Kwaah, Director of Electoral Services at Ghana’s Electoral Commission, said technology is improving voter registration, biometric verification and public access to electoral information — while acknowledging AI can be misused. “Transparency and inclusion are the surest way to deliver elections that are free, fair and acceptable.”
Bishop Emmanuel Kofi Fianu, chairman of the National Peace Council, urged cooperation on electoral violence, misinformation, political intolerance and illicit campaign financing, arguing that integrity must cover the full electoral cycle.
The detection problem underneath
The accountability question Prempeh poses is harder in African contexts for a technical reason the summit did not address.
North-West University research published this year found that translation quality directly determines how well AI misinformation detection performs in low-resource languages. Systems that struggle with isiZulu and Sepedi will struggle equally with Hausa, Twi, Dholuo and Wolof — the languages in which African political narratives frequently form before reaching English.
Meta has also been phasing out third-party fact-checking across African markets.
Assigning accountability requires first being able to establish what happened. On current capability, that remains the weaker link.
The summit, themed “Money, Technology and Civic Rights: Defending Electoral Integrity in Africa,” was organised by the African Election Observers Network with CDD-Ghana and WAEON.
