{"id":47277,"date":"2026-04-27T12:28:17","date_gmt":"2026-04-27T16:28:17","guid":{"rendered":"https:\/\/news.maryland.gov\/dnr\/?p=47277"},"modified":"2026-04-28T14:52:04","modified_gmt":"2026-04-28T18:52:04","slug":"is-that-real-a-guide-to-identifying-fake-wildlife-videos-created-with-generative-ai","status":"publish","type":"post","link":"https:\/\/news.maryland.gov\/dnr\/2026\/04\/27\/is-that-real-a-guide-to-identifying-fake-wildlife-videos-created-with-generative-ai\/","title":{"rendered":"Is That Real? A Guide to Identifying Fake Wildlife Videos Created with Generative AI"},"content":{"rendered":"

\"Example<\/a><\/p>\n

Author\u2019s Note: despite resemblance to current GenAI visuals, <\/span><\/i>this image by user sunny305<\/span><\/i><\/a> was published in 2021, prior to the AI boom, and is likely representative of the skill of digital Photoshop artists.<\/span><\/i><\/p>\n

In the head-spinning, ever-expanding world of generative AI, a particularly popular niche is being cultivated that could cause long-term damage to wildlife and the way humans interact with them: fake animal videos.<\/span><\/p>\n

Picture seeing this on your feed: It\u2019s dark, in a fenced backyard, and nocturnal animals are out to explore. In the video, a group of bunnies on the edge of a trampoline investigate the surface, venturing forward, and then, realizing the springiness of the black mesh, begin to bounce. Soon the whole group is bouncing with enthusiasm.<\/span><\/p>\n

But, there are some issues. What seemed to be seven bunnies at the beginning of the video turns into six by the time they\u2019re bouncing. They\u2019re unusually blond for wild rabbits, and their color patterns seem to disappear. In fact one of the bouncing bunnies <\/span>does<\/span><\/i> disappear! What eerie, sinister rabbits are these, who morph their shapes and flicker out of existence?<\/span><\/p>\n

This video isn\u2019t real<\/span><\/a>; it was generated by a computer in what is now known as generative artificial intelligence. Generative AI is rooted in large language models, which take in massive amounts of data to make a predictive guess on what to generate based on the examples in its model. <\/span><\/p>\n

The results of a large language model\u2019s guesses can be unexpected. Hence, rabbit ears that are reabsorbed into another rabbit\u2019s fluffy butt\u2014the model doesn\u2019t know that this shouldn\u2019t occur, only that when many rabbits are crowded together, ears unattached to a rabbit’s head may appear over the other\u2019s back. An AI <\/a><\/span>doesn\u2019t \u201cthink\u201d<\/span> on its own<\/span>\u2014it reproduces what it has seen from the data it\u2019s been given. A 2024 study found that AI systems<\/span> largely fail to recognize object permanence<\/span><\/a>, a marker of cognitive development for infants.<\/span><\/p>\n

AI-generated wildlife videos are racking up millions of impressions and being posted on social media sites by thousands of social media \u201ccreators\u201d every month. Some social media apps such as Facebook and Instagram <\/span>have posted policies<\/span><\/a> requiring users to label AI-created images and videos as made with AI, others such as X do not currently require all users to do so.<\/span><\/p>\n

However, enforcement by the social media companies is often scattershot and many users deceive others by hiding the required \u201cmade with AI\u201d lines at the very end of a line of text or don\u2019t follow the rules at all. For example, this <\/span>Instagram video of eagles stepping in wet concrete<\/span><\/a> as construction workers look on has no clear \u201cmade with AI\u201d identifier and the caption makes it seem as if it really happened, despite being posted on Instagram.<\/span><\/p>\n

When large language models make mistakes that it assumes are factual, the technical term for these mistakes are \u201c<\/span>hallucinations<\/span><\/a>.\u201d Studies have shown that <\/span>hallucinations are inevitable<\/span><\/a> with the nature of the technology, according to a 2025 study by researchers at the University of Singapore. Because a large language model must act on its own prediction to proceed, errors build on each other, and newer generative AI models are <\/span>showing more errors<\/span><\/a> rather than fewer.<\/span><\/p>\n

System Overload<\/b><\/p>\n

While it may seem innocuous, the misinformation spread by these kinds of posts causes more harm than can be seen from the surface.<\/span><\/p>\n

Cultural depictions of animals can sway the public\u2019s feelings about wildlife. The movie \u201cJaws,\u201d released 51 years ago, contributed to longstanding negative perceptions of sharks and may have influenced an uptick in the killing of sharks. Today, fake content about wild animals that proliferates on social media could give people the wrong ideas about how animals behave or how to interact with nature safely.<\/span><\/p>\n

Most of us going for a hike don\u2019t expect to encounter a possibly dangerous animal, but if we did, would we know how to respond? Even if you knew, what about your neighbor or older uncle? If they saw a video of a grizzly bear licking a kitten, would they know this is not a likely occurrence? Would a child who has grown up with feel-good fake animal videos their whole life be able to guess this is not real?<\/span><\/p>\n

The price of a generative AI video like this could be a life.<\/span><\/p>\n

Dangers like these are already apparent in the spawning of hoards of <\/span>AI-written mushroom guides<\/span><\/a>. As any forager knows, identification of edible foods in the wild is already a serious process, and no single item more so than mushrooms. North America has several deadly mushroom species that look nearly identical to edible ones and can only be told apart by the most experienced mycologists. A recent <\/span>alarming uptick in mushroom poisonings in California<\/span><\/a> underlines the danger; such poisonous fungi can kill in the space of hours and may <\/span>cause irreparable damage<\/span><\/a> to your body if you survive.<\/span><\/p>\n

It\u2019s not just safety that this kind of information can affect; our sense of awe and wonder in the world around us is an unintended casualty of the proliferation of convincing fakes. <\/span>Creativity has been shown to suffer<\/span><\/a> too as a side effect of using AI tools; why think when a <\/span>machine is doing it for you<\/span><\/a>?<\/span><\/p>\n

Reading Between the Code<\/b><\/p>\n

There are some common ways to tell whether the video you\u2019re watching is generative AI:<\/span><\/p>\n

Video length<\/b>: Most generative AI video sources can only generate 30 seconds of video at a time and have little consistency. LLMs often don\u2019t have much memory, and each prompt given to the software will result in a new video. Thus the model will slightly, or even dramatically, change the appearance of characters in the video from prompt to prompt or even from scene to scene. Some video prompters have come up with complicated workarounds for consistency, but for the time being, most accounts intentionally posting generative AI videos won\u2019t go through that amount of work.<\/span><\/p>\n

Consider the source<\/b>: Often AI accounts will post multiple versions of similar videos, with the hope that one will get views and likes, leading to monetization. Does the source have many similar videos, or does the video seem to have an agenda? Is the source a new account or one that doesn’t seem able to respond? Many of these types of accounts across platforms profit from clickbait and false information.<\/span><\/p>\n

Visual clues<\/b>: A few years ago, counting the fingers on the hand of a suspected AI photo was an easy way to tell a fake image. While large language models have gotten more complex,\u00a0 glitches do happen, as in the bunny-trampoline video. Mistakes may also be more subtle, so here are some questions to ask yourself:<\/span><\/p>\n