The standard worry about generative AI for the past three years has been that fake videos will fool voters, juries and the wider public into believing things that never happened. The fear is not unreasonable. Synthetic images and audio have improved rapidly, and the 2026 US midterms are already the first electoral cycle in which deepfakes have been deployed at industrial scale, according to a Detroit News investigation.
But a growing body of research suggests the larger threat may run the other direction. The phenomenon has a name: the liar's dividend. It describes the shift that happens once fakes are common enough that anyone can plausibly dismiss real evidence as fabricated. The burden of proof inverts.
Where the term comes from
Legal scholars Robert Chesney and Danielle Citron coined the phrase in 2019, and the academic literature has caught up since. A 2024 Yale study published in the American Political Science Review tested whether voters would believe politicians who claimed compromising material was AI-generated. The result was that claims of fakery worked best on text-based reports and were largely ineffective against video evidence, at least for now.
Real-world examples have piled up. The New York Times has documented cases during coverage of the 2026 Iran war in which authentic videos, including footage of Israeli prime minister Benjamin Netanyahu, were dismissed as AI-generated by online viewers. In Arizona, attorney general Kris Mayes warned in February that deepfake video calls were being used in romance scam operations, and that targets often disbelieved subsequent verification attempts.
Why the panic itself does harm
The Brennan Center has argued for what it calls shrinking the liar's dividend, through provenance standards, content credentials and verification workflows. Their argument is that public hysteria about deepfakes, however accurate it may be about the technology, also lowers trust in genuine media. If everything could be fake, nothing has to be proven.
None of this is to say synthetic media is harmless. Deepfake-driven scams, non-consensual imagery and political smears are real problems with real victims. The misconception is the assumption that the harm flows mostly from successful fakes. As the Yale researchers put it, claims of misinformation cost nothing to make. That asymmetry, more than any individual fake video, is what reshapes public trust.
Sources
- i. en.wikipedia.org
- ii. www.cambridge.org
- iii. www.brennancenter.org
- iv. truescreen.io
- v. eu.detroitnews.com
- vi. www.biometricupdate.com
Commentarii · 0