Brooklyn was still missing four months later, and Butler said she continues to receive fabricated images from people claiming to have found him roughly once a month. WIRED tested one of those images and found that ChatGPT verified it had generated at least one picture sent to Butler. The publication was also able to create a closely similar image using the original photo of Brooklyn and a 29-word prompt.
The same authenticity problem is affecting organizations that publish real animal imagery. We Animals, a nonprofit that distributes work from 175 photojournalists documenting alleged mistreatment at farms, circuses and research facilities, prohibits its photographers from using AI. Even so, viewers have begun questioning whether some of its footage is synthetic. The organization now expects to provide more behind-the-scenes material and additional information about how submissions are authenticated. It also plans to use technology that can preserve information about where an image originated and how it was edited.
Victoria de Martigny, We Animals’ director of visual content, said losing credibility could “open up the door to people questioning all of the work,” adding that “we don’t ever want to be in that position.”
The concern goes beyond whether a photo looks convincing. Animal welfare and conservation groups often rely on compelling visual evidence to attract attention and donations. Synthetic clips can compete with that material because they are easier and cheaper to produce, while repeated exposure to fakes can make audiences more skeptical of genuine images.
Oscar Horta, a philosopher and animal activist who recently helped direct a short film about AI and wildlife, pointed to fabricated videos showing dramatic animal rescues during disasters. He said some portrayals could give viewers a distorted impression of how real rescue work happens. “There are deepfakes of polar bears drowning and people on boats coming and rescuing them,” Horta said, describing the scenes as “ridiculous” and “unrepresentative of what it means to help animals.”
Researchers are also considering changes at the model level. Jeff Sebo, director of the Center for Mind, Ethics, and Policy at New York University, has urged AI developers to include guidance aimed at reducing outputs that could mislead people about animal suffering. The objective, Sebo said, is to “emphasize the importance of staying grounded in evidence and reason” while avoiding systems that become “overly preachy, overly moralizing, or refusing reasonable user requests.”
Regulators are pursuing a different approach through labeling requirements. New rules in California and the European Union require widely used AI image generators to embed signals indicating that content was created with AI. Social platforms are then expected to use those signals to identify synthetic images and videos for users.
Some verification tools are already available. ChatGPT, Gemini and Meta AI can determine whether particular images were generated by their own systems. But those checks generally require people to upload media manually, and usage limits can make the process impractical for someone encountering large amounts of questionable content.
That leaves much of the burden on users to inspect images themselves or decide whether a post looks suspicious. For people like Butler, the effect is cumulative. She now blocks accounts that post AI-generated cat videos and spends more time in an online community for artists who oppose AI, where she feels more confident that the work she sees was made by people.
The broader challenge is no longer simply detecting individual fakes. As synthetic animal content becomes more common, legitimate creators and organizations are increasingly being asked to prove that real images are real, while users face a growing verification burden before trusting what appears in their feeds.
This analysis is based on reporting from Wired.
Image courtesy of Fox 35 Orlando.
This article was generated with AI assistance and reviewed for accuracy and quality.
