Recent advancements in AI tools have raised alarms regarding the creation of misleading images, particularly involving controversial figures such as Jeffrey Epstein. A recent study highlights how easily these tools can fabricate realistic images linking prominent politicians to the convicted sex offender, which highlights the growing concern over misinformation in the digital age.
Social media has become a breeding ground for AI-generated images that falsely depict Epstein socializing with various politicians, including New York Mayor Zohran Mamdani and award-winning filmmaker Mira Nair, as noted by AFP’s fact-checkers.
A recent study conducted by the US disinformation watchdog, NewsGuard, tasked three leading image generators with creating images of Epstein alongside five notable politicians, including former President Donald Trump, Israeli Prime Minister Benjamin Netanyahu, and French President Emmanuel Macron.
The study found that Grok Imagine, a tool created by Elon Musk’s xAI, was capable of producing “convincing fakes in seconds.” Among the generated images was a strikingly lifelike depiction of a younger Trump and Epstein with girls. While Trump has been documented with Epstein at a number of social functions, no publicly available image exists featuring them alongside minors.
Google’s Gemini took a different approach; it refrained from creating an image of Epstein with Trump but successfully produced realistic representations of the late sex offender with Netanyahu, Macron, Ukrainian President Volodymyr Zelensky, and UK Prime Minister Keir Starmer. These fabricated images suggested that Epstein was mingling with these politicians at parties, on private jets, and relaxing on beaches.
The findings underscore the ease with which malicious actors can exploit AI imaging tools to churn out viral fakes that appear remarkably authentic. NewsGuard remarked on the pervasive nature of this issue, stating, “fake images have become so routine that it’s difficult to distinguish real photos from AI-generated ones.”
Interestingly, when prompted, OpenAI’s ChatGPT opted not to generate any images of Epstein with the politicians, citing ethical considerations. The model stated it would not create images featuring real individuals in scenarios that implicate sexual abuse or involve minors.
Detecting Fakes
No immediate response was received from xAI regarding the findings. Meanwhile, researchers analyzing the deceptive images linking Epstein with Mamdani and Nair—images that amassed millions of views on X—discovered a SynthID watermark embedded in the content. This invisible marker is designed to signal that the images were generated using Google’s AI technology. A Google spokesperson confirmed to AFP that they include an imperceptible SynthID watermark to help users identify AI-generated content.
This study is particularly timely, as the Justice Department recently released a trove of Epstein-related files, consisting of over three million documents, photos, and videos, related to its investigation. Epstein died by suicide while in custody in 2019, and the case continues to ensnare high-profile individuals worldwide, including former British Prince Andrew, noted American intellectual Noam Chomsky, and Norway’s Crown Princess Mette-Marit.
However, the ongoing discourse surrounding Epstein has fueled a surge in disinformation. This week, a fabricated social media post attributed to Trump circulated widely, asserting that he would eliminate all tariffs against Canada if Prime Minister Mark Carney confessed to involvement with Epstein. An AFP review of the materials does not substantiate any claims of Carney’s involvement in Epstein’s alleged criminal activities.
As we navigate the complexities of modern technology and misinformation, it becomes essential to maintain a critical eye towards the content we encounter online. Recognizing the potential for deception is crucial in the fight against disinformation.
