Deepfakes and election integrity, research summary

Synthetic media in political contexts has been the subject of sustained research since the last US presidential cycle, and the 2024 global election year produced a wealth of new case studies. Organizations including the Oxford Internet Institute, the Stanford Internet Observatory, and the Reuters Institute for the Study of Journalism have documented incidents from multiple countries, along with platform responses and voter reactions. The cumulative picture is more complicated than the early framing that deepfakes would decisively swing elections.

Some patterns have emerged. Audio deepfakes, because they are cheaper to produce and harder to detect visually, appeared more often than full video forgeries. Synthetic content frequently targeted candidates late in campaigns, when corrections had less time to reach voters. Authentic material taken out of context continued to outperform purely synthetic content in reach, a reminder that traditional misinformation techniques have not been replaced by new ones. Platform labeling and rapid response by fact-checking networks mitigated some but not all of the damage.

Government responses have varied. The European Union's Digital Services Act and accompanying risk assessments require large platforms to address election-related manipulation, including synthetic media. US states have enacted narrower laws requiring disclosure of synthetic political advertising in defined windows before elections. India, Brazil, and several European countries held national elections under new or updated rules and produced useful after-action reports that researchers and regulators in other jurisdictions are now reading.

Editor's note: The responsible framing is neither that deepfakes will destroy democracy nor that they are a non-issue. They add a new tool to an old toolkit of political deception, and they interact with existing media ecosystems in ways that depend heavily on local context. Voters, journalists, and election officials benefit from specific protocols, such as agreed verification channels for candidates, rather than from generic alarm.