Google's Veo 3 AI creates convincing deepfake videos of election fraud
Google's Veo 3 text-to-video AI system creates highly realistic deepfake videos that can depict election fraud, riots and other controversial events. A TIME investigation demonstrated the tool's ability to produce convincing misinformation, and researchers warn that inadequate safeguards could enable the spread of false videos that threaten democratic processes and public safety.
What happened
TIME used Google's Veo 3 AI to create fabricated video clips depicting riots, election fraud, and conflict scenarios. Videos created included a Pakistani crowd setting fire to a Hindu temple, Chinese researchers handling a bat in a wet lab, an election worker shredding ballots, and Palestinians accepting U.S. aid in Gaza.
After release, users created and shared multiple fake videos including false news about public figures, such as a news anchor announcing the death of J.K. Rowling, and violent events. Though the videos contained some inaccuracies, they were realistic enough to potentially fuel social unrest or violence if misused.
Reported impact
Experts warn that Veo 3 and similar AI video tools could amplify misinformation, propaganda, and societal division by making it very difficult to distinguish real from fake videos, threatening public safety, democratic processes, and individual reputations.
Index analysis
Veo 3 produces much more convincing videos than earlier text-to-video generators, with dialogue, sound effects, and realistic physics. While Google implemented safety measures such as blocking harmful prompts and adding visible and invisible watermarks, these safeguards were limited and easily bypassed, such as by cropping out the visible watermark.
Minimal prompting led to provocative and misleading content, and users could combine clips to fabricate complex fake news stories. The technology's sophistication reduces traditional visual cues that videos are fake, making verification harder.
Governance controls
Google implemented safety measures including blocking specific harmful prompts and adding both visible and invisible watermarks to generated videos. The tool was made available at a monthly subscription cost and was subject to pre-release testing to reduce harmful outputs.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.