AI models can identify the city and country of a photo without any metadata with 87% to 91% accuracy [1].
This capability creates significant privacy risks for social media users who believe removing location tags or GPS data secures their identity. By analyzing visual cues, AI can bypass traditional privacy settings to track individuals.
Security firm McAfee conducted the study using more than 21,000 travel photos from around the world [1]. The research demonstrates that AI does not need embedded data to determine a location. Instead, the models scan for visual markers such as specific buildings, road markings, and shop signs [1].
Other visual indicators include skylines and food stalls, which provide enough context for the AI to infer a geographic position [1]. This process allows AI to track locations from social media posts about nine out of 10 times [2].
The study suggests that the sheer volume of visual data available in public posts makes it easier for AI to build accurate location profiles. Even images that appear generic to a human observer may contain distinct architectural or environmental clues that AI can recognize instantly [1].
As these models become more accessible, the ability to deanonymize a user's location increases. The findings highlight a gap between user perception of privacy and the actual technical capabilities of modern image recognition software [1].
“AI models can identify the city and country of a photo without any metadata.”
This study indicates that metadata stripping—a common privacy recommendation—is no longer a sufficient defense against location tracking. As AI improves its ability to recognize 'visual fingerprints' of cities and landmarks, the risk of digital stalking and unauthorized surveillance increases for anyone posting travel imagery online.
