AI-Generated Images of Pope Francis Drive Traffic to Network of Suspicious Websites in Coordinated Campaign

In our ongoing efforts to monitor coordinated accounts discovered by the Vera AI alerts — a workflow designed to monitor, detect, and iteratively update lists of coordinated social media accounts by analyzing behavioral patterns and interaction metrics — we (a team of researchers at the University of Urbino) recently uncovered a significant disinformation campaign on Facebook pertaining to Pope Francis’ health. This operation generated over 400,000 posts a week, beginning on February 23, 2025. The campaign capitalized on the widespread attention surrounding the hospitalization of the Pope, leveraging public interest and concern to disseminate misleading content.

The campaign's posts, while numerous, were strikingly uniform in their content. Each post contained identical text, accompanied by a visual collage (Figure 1) comprising both AI-generated and authentic images of the Pope. These collages were likely created to circumvent automatic detection systems. 

Although one specific image within these collages was confirmed as AI-generated by the elDetector project of Univision Noticias, their fact-check label has only been applied to a small fraction of the total posts.

Additionally, thanks to support from AFP, we located an early version of one of the AI images featured in these collages that still display the Grok xAi logo (Figure 2).

The campaign's strategy involved posting a link in the first comment of each content, typically by the same user who created the original post (Figure 3).

The links below direct users to a network of similar-looking breaking news blogs, all sharing the same template and tracking scripts (Table 1).

       Table 1. A sample of website domains in the network (link to urlscan.io analysis)

Operations like these demonstrate how easily Meta's Facebook policies against false news, synthetic images, and links to potentially harmful websites can be circumvented. This case also reveals the limitations of Meta's AI systems in effectively identifying near-duplicate versions of previously fact-checked images. These findings are particularly concerning given the phasing out of the third-party fact-checking programme in the United States, leading to non-detection of harmful posts which can easily exploit vulnerable users of the platforms.

Authors: Fabio Giglietto, Giada Marino, Anwesha Chakraborty, Massimo Terenzi (University of Urbino)

Special thanks to Denis Teyssou (AFP) and Francesco Poldi (EU Disinfo Lab)

Editor: Jochen Spangenberg (DW)