Deepfakes Are Already on Kenya's 2027 Ballot. Plan for It.

In this cycle a fabricated, AI-generated image of Communications Authority Director-General David Mugonyi apparently defending a media shutdown circulated in Kenya. Africa Check has separately debunked a fake "Infotrak poll" graphic showing President Ruto in third place.

Neither was sophisticated. Both were shared widely by people who believed them.

What is actually documented

Researchers analysing Kenyan online disinformation have identified 17 major coordinated campaigns, with significant overlap between the accounts involved — meaning the same networks running multiple operations. The campaigns feature smear content and ethnically charged hashtags, and analysis found the most sophisticated among them focused on dismantling the political dominance of the Mount Kenya region.

TikTok has been identified as the leading disseminator of misinformation and disinformation in Kenya.

Kenya's National Intelligence Service Director-General has warned that malicious actors, both domestic and foreign, are weaponising social media and artificial intelligence. Cabinet Secretary Kabogo has warned media to prepare for AI, fake news and misinformation ahead of 2027. The UN's Independent International Scientific Panel on Artificial Intelligence has warned that AI-generated misinformation and deepfake video threaten public trust and election integrity.

So this is not a prediction. It is a description.

What a campaign should assume

Four things, and planning against them costs almost nothing.

You will be deepfaked. Audio is the likeliest vector, because it is cheapest to produce, hardest to disprove and travels best on WhatsApp. A thirty-second clip of "you" saying something inflammatory about another community is the highest-damage, lowest-cost attack available.

It will arrive at the worst moment. Late in the campaign, ideally after campaigning has legally ended on 7 August 2027, when your ability to respond at scale is constrained.

It will spread in closed networks first. WhatsApp groups, not public timelines. By the time it is visible to you it has been circulating for days.

A correction will not catch it. It never does. The goal of a response is not to catch up; it is to give your own supporters something to say.

The plan, in advance

Nominate a response person now. One named person who monitors and responds. Not the candidate. Not "the comms team".

Build the channel before you need it. A response reaches people through the structure you already have — ward coordinators, volunteers, supporters with your number saved. A campaign that has been in weekly contact with 400 volunteers for a year can put a correction into 400 real conversations in an hour. A campaign that has a Facebook page cannot.

This is the single most underrated reason to build a real supporter structure early. Its value on the day a fake circulates is enormous, and it cannot be assembled in the moment.

Pre-record your own reference material. Ironically useful: a body of genuine video and audio of the candidate makes a fake easier to contrast against.

Write the holding response now. Two sentences, ready to send. "That recording is not me. Here is where I actually was, and here is what I actually said." Drafting it in the moment costs you the first two hours, which are the only ones that matter.

Report it. To the platform, to the IEBC, and to fact-checkers. Slow, but it builds the record that later matters.

And do not become the problem

The Electoral Code of Conduct has been in force since 20 August 2026, and a party and its candidate can be answerable for what supporters do on their behalf.

A campaign that runs anonymous accounts, or tolerates supporters doing so, is one screenshot away from a story that is much worse than whatever the accounts were posting. Tell your people, in writing, what your campaign does not do — and be able to show who is acting for you, which requires a record of who your people are.

Related: what the Code of Conduct now binds you to.