Artificial intelligence

AI-generated video of glass bridge breaking misleads online

Copyright © AFP 2017-2025. Any commercial use of this content requires a subscription. Click here to find out more.

Glass bridges and viewing platforms are popular attractions across China, but a video viewed more than two million times across multiple social media posts of one such bridge collapsing is a fabrication. The clip contains visual inconsistencies such as people blending into the background or each other, hallmarks of videos created using AI. The last reported incident of a glass bridge breaking in China was in May 2021.

"A viral glass tourist bridge in China cracked and collapsed," read Indonesian-language sticker text over a video shared on Instagram on May 20, 2025.

The clip appears to show people crossing a glass platform against a cliff face, which breaks and falls into the ravine below.

Screenshot of false post, taken on May 23, 2025

The clip, which also spread on Instagram, YouTube, Tiktok and Facebook, garnered more than two million views.

Comments from some users indicate they believed the footage was real.

"The load was too much," one user commented.

"Terrifying, isn't it?" wrote another.

China is believed to have more than 2,000 glass bridges and platforms at various tourist sites across the country (archived link).

However, the last reported safety incident at a glass bridge in China was in May 2021 -- a man was left dangling at a tourist spot in northeastern Jilin province when the glass panels were damaged by high winds, according to a report by state news media Xinhua (archived link).

A close inspection of the clip also shows inconsistencies typical of AI-generated videos. 

These include people appearing to melt into the cliff before reappearing in different coloured clothes, while two of the people appear to combine into one person after the bridge collapses. One section of the glass guardrail also disappears immediately after the bridge breaks.

Despite the rapid progress of generative AI, errors still appear in AI-generated content. These flaws are often the clearest signs of a fabricated image.

Screenshot of inconsistencies on false post higlighted by AFP

Jacobo Castellanos, coordinator for human rights organisation WITNESS, which developed a Deepfakes Rapid Response Force said "it's clear that the video in question is AI-generated" (archived link). 

"Several signs point to its synthetic nature, including people morphing or disappearing, as well as various unnatural movements and transitions that are characteristic of the current capacity of generative video tools," he said. 

AFP has fact-checked AI-generated images before here. 

Is there content that you would like AFP to fact-check? Get in touch.

Contact us

Previous article