A surge in AI-generated videos depicting bear encounters is exacerbating public anxiety in Japan, where a record number of fatal attacks has already heightened tensions. Experts warn that the realistic clips, many produced using OpenAI's video generator Sora, could undermine safety messaging and put people at greater risk.
Two of the most notorious fake clips show bears destroying solar panels and carrying off a pet dog. Others depict a newsreader reporting sightings, an older woman feeding a bear, and even a schoolgirl fighting off an animal—a dangerous response given that black bears can weigh up to 130kg and Ussuri brown bears up to 400kg, easily outrunning humans.
A search by the Yomiuri Shimbun on TikTok found that about 60% of 100 selected clips were fake. Some included realistic details such as references to locations where sightings have occurred, leading to confusion. In one case, an AI-generated 'news report' showed a bear being captured after entering a convenience store in Noshiro, Akita prefecture, but officials confirmed no such incident occurred.
Experts are particularly concerned about clips showing people feeding bears, despite repeated warnings not to carry food while hiking. Shinsuke Koike, a professor at Tokyo University of Agriculture and Technology, said feeding bears is extremely dangerous and could cause them to lose their fear of humans.
The number of fake videos has soared as users capitalise on public interest amid a record 13 fatal bear attacks this year—more than double the previous high—and over 100 injuries. About 20,700 sightings were reported between April and September, 7,000 more than in the same period in 2024.
Disruption from real encounters is significant: Japan Post suspended deliveries in parts of Akita prefecture, where self-defence forces have been deployed to help trap bears. NTT East is reviewing safety protocols for engineers, and logistics firm Yamato may suspend parcel services if bear risks are deemed high.
