Maduro Capture Triggers Flood of Deepfakes, Exposing Gaps in AI Safeguards

Overview

• NewsGuard tracked five fabricated images and two miscontextualized videos tied to the capture, drawing more than 14.1 million views on X in under two days.
• G1’s Fato ou Fake used Google’s SynthID to determine that a widely shared post purporting to show Maduro escorted by U.S. agents was created with Google’s AI.
• The New York Times tested a dozen popular image generators and found most—including Gemini, OpenAI-based tools and X’s Grok—would produce staged capture images within seconds.
• OpenAI said it employs safeguards for public figures, yet tests showed third‑party sites using its model still generated the prohibited images even when ChatGPT itself refused.
• Fact-checkers determined a viral clip of people toppling a giant poster was from July 29, 2024 election protests in Maracay rather than a new celebration of the capture.