Analysis — Sub-Saharan Africa11/15/2026

Beyond Deepfakes: AI, the ‘Liar’s Dividend’ and the Crisis of Political Evidence in Zambia

Zambia’s 2026 election exposed a crisis of political evidence: AI can fabricate events, while allegations of AI manipulation can also be used to dispute authentic recordings.

Written by Ollus Ndomu
Analysis — Sub-Saharan Africa11/15/2026

Beyond Deepfakes: AI, the ‘Liar’s Dividend’ and the Crisis of Political Evidence in Zambia

Zambia’s 2026 election exposed a crisis of political evidence: AI can fabricate events, while allegations of AI manipulation can also be used to dispute authentic recordings.

Written by Ollus Ndomu

Conceptual illustration of deepfake technology and facial recognition risks.

Conceptual illustration of deepfake technology and facial recognition risks.


Beyond Deepfakes: AI, the ‘Liar’s Dividend’ and the Crisis of Political Evidence in Zambia

Zambia's 2026 election exposed a crisis of political evidence: AI can fabricate events, while allegations of AI manipulation can also be used to dispute authentic recordings.

Zambia's August 2026 general election exposed a troubling shift in political communication. Artificial intelligence was no longer merely a tool for manufacturing falsehoods; it had become a convenient explanation for denying uncomfortable truths. In a deeply contested political environment, the authenticity of evidence increasingly risked becoming a matter of partisan allegiance rather than independent verification. The danger extends beyond misleading voters. When political actors can dispute genuine recordings by invoking AI, accountability itself becomes negotiable.

Political researchers describe this phenomenon as the “liar's dividend”: the advantage individuals may gain by dismissing authentic but damaging information as artificially generated. Its significance lies less in whether every denial succeeds than in the uncertainty it creates. A politician need not prove that a recording is fabricated. Sometimes, merely suggesting that it could be is enough to shift public attention from the substance of the evidence to arguments about its authenticity. Zambia's election provided a revealing illustration of this political calculation.

On 9 August, opposition politician Binwell Mpundu posted a Facebook video declaring his willingness to “die” or “kill” over the election. Following criticism, he deleted the recording and claimed it was AI-generated, alleging that his account had been compromised. Opposition presidential candidate, Brian Mundubile, subsequently defended him, encouraging the public to disregard the video. However, the explanation unraveled when Mpundu acknowledged making the remarks. “The truth is this is the statement I made,” he said, according to Lusaka Times on 11 August. He maintained that his words had been taken out of context.

The contradiction revealed more than an unsuccessful attempt at damage control. Mpundu's initial explanation gave his political allies a basis for defending him without first establishing the recording's authenticity. Mundubile's intervention demonstrated how an unverified denial could acquire political legitimacy through endorsement by a senior opposition figure. When Mpundu reversed his position, the episode exposed the danger of political loyalty preceding factual scrutiny. The Electoral Commission of Zambia (ECZ) subsequently revoked his accreditation to the National Results Centre, citing public safety concerns in a statement dated 11 August. The denial had not secured immunity. Nevertheless, it illustrated how AI can be invoked to contest responsibility before institutions establish the facts.

This creates a dilemma for electoral governance. Authorities must respond to conduct that threatens public safety, but their interventions occur within an environment where institutional decisions are themselves politically contested. An AI allegation can therefore serve two purposes: challenge the authenticity of damaging evidence and provide supporters with grounds to question subsequent enforcement. Neither outcome proves that citizens accepted Mpundu's explanation. Rather, the episode demonstrates how the burden of establishing truth can become entangled with political allegiance. Even a failed denial can expose weaknesses in the relationship between evidence, leadership and public trust.

The wider election demonstrated the opposite danger: fabricated information acquiring the appearance of political truth. On 23 July, iVerify Zambia rejected a viral assassination story concerning Mundubile, citing police verification and indicators of AI-generated storytelling. On 14 August, while votes were still being counted, the organisation challenged premature online claims that Mundubile had won the presidency. The ECZ had made no such declaration. Later, a synthetic video posted on 20 August purported to show ECZ Chairperson, Mwangala Zaloumis, conspiring with security chiefs to manipulate the election. These incidents were not identical, nor was every misleading claim necessarily AI-generated. Collectively, however, they illustrated how fabricated and unverified political narratives could be presented as credible evidence, creating opportunities to amplify suspicion without establishing their actual influence on voters.

The deeper institutional problem concerns the conditions under which such narratives become persuasive. Where citizens cannot readily examine official electoral records, distinguish polling-station figures from constituency totals or obtain timely explanations of disputed results, unofficial claims may fill the information gap. But dismissing every inconsistency as disinformation would be equally dangerous. Genuine electoral concerns deserve investigation. Fabricated evidence demands exposure. The distinction is fundamental. Fact-checkers such as iVerify can establish whether particular claims are credible, but verification alone cannot guarantee that corrections will reach the same audiences or overcome entrenched political loyalties. Electoral transparency and independent scrutiny must therefore reinforce technological verification.

Zambia's experience has wider implications for Southern Africa. Ahead of and during Namibia's November 2024 elections, Namibia Fact Check documented deceptive AI-generated political material, while a March 2025 report produced by Namibia Fact Check, a project of the Institute for Public Policy Research, identified coordinated disinformation campaigns, weaknesses in electoral information integrity and the limited preparedness of electoral authorities to counter such threats. The comparison reveals a regional challenge extending beyond individual deepfakes. Electoral authorities must increasingly defend both the credibility of official results and the authenticity of the evidence informing public debate. This requires accessible electoral records, timely communication, independent verification and stronger digital literacy. Otherwise, political actors can exploit uncertainty about whether evidence is fabricated or genuine.

The central lesson from Zambia is therefore not that AI determines election outcomes or guarantees impunity. It is that technology has expanded the political opportunities created by uncertainty. False evidence can manufacture suspicion; genuine evidence can be dismissed as false. When both tactics operate within an already polarised environment, democratic accountability depends not only on establishing what happened, but on sustaining public confidence in how the truth is established.