AI-generated video falsely linked to deadly landslides in Nepal

約 5 分鐘閱讀

站內可閱讀至 (台北時間)

本站保存版本
字級

"Terrifying scenes of a landslide in the mountains of Nepal," reads an X post from September 29, 2026, in simplified Chinese.

It retweeted an Arabic-language post which includes a video showing a massive landslide cascading down a mountainside next to a white multi-storey building, sweeping away cars and trees in its path.

The clip was shared with similar claims across different social media platforms in multiple languages, including English, Hindi, Spanish and French.

Heavy rainfall triggered floods and landslides across northern India and neighbouring Nepal in late September 2026, leaving at least 68 people dead (archived link).

The Himalayan nation, where a devastating flood in late August killed more than 1,400 people and left thousands more missing, was hit by new flooding and landslides that killed at least 12 people.

Scientists have warned in recent years of an increase in extreme weather events across South Asia, linking them to rising temperatures and changing climate patterns.

But the video is AI-generated.

An analysis of the footage identified several visual anomalies that are hallmarks of AI-generated content.

The building's structure is blurred and distorted, and the flow of the debris making its way downhill is unnatural.

The powerful mudslide can be seen sweeping away cars and trees on the roads below, yet the figures on the left remain motionless without attempting to flee. Some poles along the side of the roads are seemingly unaffected by the massive landslide.

The video also shows debris falling at high speed from the very edge of a multi-storey building, yet the building remains unscathed.

High-rise constructions as seen in the video are not a common feature of Nepal's mountainous districts, where settlements are typically low-rise due to the steep terrain and limited flat land (archived here and here).

An analysis of the video using Hive Moderation, an AI detection tool, found it 99.4 percent "likely to contain AI-generated or deepfake content" (archived link).

AFP has debunked other AI-generated misinformation.