OpenAI's latest AI models, o3 and o4-mini, have taken the tech world by storm with their impressive image-analysis capabilities. However, this new feature has also raised significant privacy concerns, as users have discovered that the models can be used to deduce locations from uploaded images, potentially leading to unwanted revelations about individuals' personal lives.

The models' ability to "reason" through images allows them to crop, rotate, and zoom in on photos, even blurry and distorted ones, to thoroughly analyze them. When combined with their web search capabilities, this feature makes for a potent location-finding tool. Users on social media platform X have been experimenting with the models, feeding them images of restaurant menus, neighborhood snaps, and self-portraits, and instructing o3 to play a game of "GeoGuessr," an online game that challenges players to guess locations from Google Street View images.

The implications of this technology are far-reaching and concerning. With o3's capabilities, a bad actor could potentially screenshot someone's social media post and use ChatGPT to try to doxx them, revealing their location and compromising their privacy. While this is not a new concern, the ease and accuracy with which o3 can deduce locations make it a particularly pressing issue.

TechCrunch conducted its own tests, comparing the location-guessing skills of o3 and an older model, GPT-4o, without image-reasoning capabilities. Surprisingly, GPT-4o arrived at the same, correct answer as o3 more often than not, and took less time. However, o3 was able to correctly identify a location that GPT-4o couldn't, highlighting its superior capabilities.

Despite its impressive abilities, o3 is not infallible. Several tests failed, with o3 getting stuck in a loop or volunteering a wrong location. Users on X also noted that o3 can be pretty far off in its location deductions. Nevertheless, the trend illustrates some of the emerging risks presented by more capable, so-called reasoning AI models.

What's concerning is that there appear to be few safeguards in place to prevent this sort of "reverse location lookup" in ChatGPT. OpenAI's safety report for o3 and o4-mini does not address this issue, leaving many to wonder about the company's commitment to protecting users' privacy. We've reached out to OpenAI for comment and will update this piece if they respond.

As AI models continue to evolve and become more sophisticated, it's essential that developers and policymakers prioritize privacy and security. The potential consequences of unchecked AI capabilities are too great to ignore. As we move forward, it's crucial that we strike a balance between innovation and responsibility, ensuring that these powerful tools are used for the greater good, not to compromise our privacy and security.