ChatGPT’s capability of doing 'reverse location search' from photos raises concerns

OpenAI's recently launched AI models, o3 and o4-mini, can analyse uploaded images, crop, rotate and zoom in on them, and search the web to determine the location.

As a result, a new trend has emerged where people are using ChatGPT to identify locations shown in pictures.

Users on social media have discovered that the o3 model is particularly effective at deducing cities, landmarks, and even specific restaurants and bars from subtle visual clues.

This has raised concerns about potential privacy issues, as people could use this feature to identify individuals or locations from social media posts.

In recent tests conducted, o3 and an older model, GPT-4o, were compared for their location-guessing skills.

Surprisingly, GPT-4o often arrived at the same correct answer as o3 and took less time. However, o3 did outperform GPT-4o in one instance, correctly identifying a bar in Williamsburg from a photo of a purple rhino head.

While o3 is not flawless and can get stuck in loops or provide incorrect answers, the trend has highlighted emerging risks associated with more capable AI models.

There appear to be few safeguards in place to prevent reverse location lookups in ChatGPT, and OpenAI's safety report for o3 and o4-mini does not address this issue.

The potential for misuse of this feature is significant, and it has raised questions about the need for more robust safeguards to protect user privacy.