Nepal flood misinformation surges as AI fakes spread online

Nepal is battling a surge of misinformation alongside rescue and relief work after deadly flash floods and mudslides, as fake and AI-generated images and videos flooded social media. The false material has complicated efforts to account for missing people and understand the scale of the disaster. The floods struck on 26 August and devastated towns and villages in northern and central Nepal after what authorities believed was a collapse or avalanche of ice and rock near the Nepal-Tibet border. Floodwaters carried by the Bhotekoshi River swept away homes, vehicles, roads and bridges, and damaged hydropower projects. One report said authorities put the death toll at over 900, with more than 4,245 people still missing. The Nepal Press Council opened a dedicated complaints portal, titled "Special Complaints on the Bhotekoshi Incident", for the public to report misinformation and graphic material tied to the disaster. By 5 pm on Saturday, the portal had received 101 complaints, including complaints against 67 personal accounts that posted gruesome visuals. Council chairman Umesh Shrestha said 72 social media users had been instructed to remove inappropriate content over the past few days, most of it posted through personal accounts. He also said complaints had been lodged against some online news portals and their social media pages. The council said mainstream media largely avoided publishing misleading information and excessively graphic images. Some of the most widely shared material was quickly traced to elsewhere. A video that claimed to show Home Minister Sudan Gurung inside a tunnel helping with rescue work was actually footage from a tunnel rescue operation in Uttarakhand, India, first posted by the Fire Service Uttarakhand Police on 14 August. Another X video, posted on 27 August and showing what looked like a catastrophic flood, drew more than seven million views. Hive Moderation judged it 97.2 per cent likely to have been generated with artificial intelligence, and X’s community notes flagged it as fabricated. A separate clip purporting to show a bridge collapsing during the Nepal-Tibet floods was also assessed as entirely AI-generated. Other false visuals recycled older disaster footage from abroad. Nepal Fact Check traced a video claiming to show a glacial lake outburst flood in Rasuwa to Columbia Glacier in Alaska, while the Australian Associated Press fact-checking unit found that "before and after" photographs of a Rasuwa settlement were fabricated. A separate Yahoo report said Lead Stories’ OpenAI Verify tool found SynthID watermarks in Facebook and Instagram "before and after" images that did not show Trishuli Bazaar. Chetana Kunwar, assistant editor at Nepal Fact Check, said AI-generated content and manipulated "before and after" visuals made false impressions much easier to create and harder for the public to detect. She added that videos from disasters in other countries were being deliberately circulated as if they were from Nepal. A CyberPeace fact check said a viral landslide clip was in fact a compilation of two old videos, one from Myanmar and one from the United States. The Social Media Regulation and Management Committee, which has been tasked with dealing with graphic content, AI-generated misinformation, fake QR codes and other online deception, has begun discussions with social media platforms. Fact-checkers said the growing use of AI has added a new dimension to the longstanding problem of recycled and falsely attributed disaster footage.