Study: AI can pinpoint the location where a photo was taken with up to 91% accuracy

McAfee, a cybersecurity firm, revealed that certain artificial intelligence models can now pinpoint the geographic location where photos were taken with up to 91% accuracy, even in the absence of GPS data or any embedded image metadata. These findings emerged after the McAfee Labs team tested over 21,000 publicly available images, aiming to assess the ability of computer vision models to infer locations based solely on visual elements.

The Qwen3 VL 30B model achieved the highest accuracy rate at 91%, while the Gemma3 27B model reached 87% accuracy in identifying the city and country where the photos were taken. The study showed that architectural landmarks, signage, architectural styles, and cultural elements help models determine locations with high precision, potentially providing scammers with information to craft more convincing phishing messages.

McAfee recommended delaying the posting of travel photos until after returning home, restricting account privacy settings on social media platforms, and avoiding interaction with messages that exploit user location data. It also emphasized the importance of directly contacting official authorities to verify any messages requesting personal or financial information.