Skip to content
Submit incident
AII-1001-96153
ReportedReported by fewer than three independent publishers, with no official finding.Supporting sources are pending verification.

Google Veo 3 video generator bypasses safeguards to create deepfakes

Google launched Veo 3, a video generation tool, with safeguards against misuse, but users have reportedly discovered methods to bypass filters and create realistic deepfakes for potentially deceptive political and news content.

25 May 2025 Β· Google Β· Misinformation and synthetic media
25 May 2025
AII-1001-96153
Quick facts
Organization
Google (subject)
Category
Misinformation and synthetic media
Country
United States
Date
25 May 2025
Drafted by
Raidu Incident Detection

What happened

Google launched Veo 3, an AI video generator announced at Google I/O 2025, which creates 4K videos with synchronized dialogue and sound from text descriptions. The system includes safeguards such as automated watermarks and blocking of problematic requests, but users have reportedly bypassed these filters to generate deepfakes.

Examples shared on social media include a TV news report of a fire on Seattle's Space Needle, an announcement of a U.S. Secretary of Defense's death, and recreations of movie scenes, with users noting the videos are nearly indistinguishable from real content. A Verge journalist discovered that the model's safety filters are not very effective and the system generates dialogue not included in the original text prompt.

Reported impact

The realistic quality of Veo 3-generated videos raises concerns about their use to create deceptive political content and false news. Google acknowledged ethical risks related to the potential for misuse to create fake news in political contexts.

Index analysis

The incident demonstrates a gap between stated safety measures and actual system behavior. Despite Google's implementation of content filters and watermarking, users were able to generate realistic deepfake videos that evade the intended safeguards.

Governance controls

Google implemented automated watermarking of generated content and blocking mechanisms for problematic requests. Access to Veo 3 is initially limited to U.S. users through monitored environments and requires a subscription to Google AI Ultra at $249.99 per month.

Classification

Organization
Google (subject)
Category
Misinformation and synthetic media
Industry
Media, social platforms and marketing
Country
United States
Harm status
Potential harm
Intent
Unintentional
Timing
Post-deployment
Harm types
Public interest
Harmed
General public
Failure modes
Deepfake video, Content moderation failure
AI systems
Veo 3

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.

How sources are verifiedSuggest a source for this incident