ChatGPT's New Image Analysis Capability Raises Privacy Concerns

This ChatGPT trick can reveal where your photo was taken - and it's unsettling

OpenAI's latest AI models, o3 and o4-mini, can analyze images beyond just recognizing objects; they can zoom in, crop, and detect visual clues in photos to help identify places, landmarks, and even specific businesses. This capability is fascinating, but it's also raised some privacy concerns about how easily people can use AI to reverse-engineer location data from images. Also: ChatGPT just made it easy to find and edit all the AI images you've ever generated TechCrunch reported that people are uploading photos -- anything from restaurant menu snapshots to casual selfies -- and asking ChatGPT to guess where they were taken. The AI does this by looking at everything in the image: the type of buildings, landscape features, and even subtle hints like the architecture or the layout of a city. It then taps into its knowledge database and sometimes the internet to make an educated guess about the location. The results can be surprisingly accurate, and it's got people talking. Some users have even tested the AI with blurry photos or images with partial objects, and the model still managed to make a guess. Also: Is ChatGPT Plus worth your $20? Here's how it compares to Free and Pro plans I love playing GeoGuessr, a game where you're dropped into a random location on Google Street View and have to guess where you are based on the surroundings. So this ChatGPT capability was super fun to test myself. I uploaded a few photos to see how ChatGPT would handle the challenge. However, as with any AI, the result was not always perfect. The first photo was a picture of my house. ChatGPT immediately guessed the location, though not based on metadata. Instead, it made an educated guess by analyzing the glacier boulder in the side yard, the types of trees visible in the photo, and the wood siding on the houses around me. It mentioned that this combination of natural features and architectural styles was typical of the region I had previously mentioned in a conversation -- where I'm from. It was accurate but still relied on some prior context, which made me wonder how much of this was due to our earlier conversation. For the second photo, I uploaded an image of a building with part of an office sign visible. ChatGPT quickly zoomed in on the sign and used the partial information to narrow down the location. It didn't take long before it deduced the city and even the area where the building was located. I was impressed with how the model used such a small visual clue to make a fairly precise guess. Also: How to use ChatGPT: A beginner's guide to the most popular AI chatbot For my third test, I gave it a more challenging photo: a scenic view with no obvious landmarks. ChatGPT explained that while it could analyze the image for visual clues, it needed distinctive landmarks or location tags to make a confident guess. It also mentioned that without GPS data or EXIF tags, it could only read the scene itself -- trees, buildings, terrain -- and would be less certain in its guess. I played with ChatGPT a few more times after this and noticed the AI starting to get more stubborn about not identifying locations in my photos. It'd tell me that it would try to narrow down the options, but it needed more data to go on to give me an accurate answer. Still, one of the main reasons people seem drawn to this trend is because ChatGPT does analyze more than just the obvious features of a photo. Even without metadata like EXIF data or other direct data extraction, the AI interprets visual clues. This experiment made me realize how powerful, and sometimes eerie, these AI models can be when it comes to "reverse location searches" from photos -- even if they aren't great at it yet. It also made me think about the privacy concerns. Also: OpenAI's most impressive move has nothing to do with AI While AI tools can be fun and useful at times, it's important to be aware of the risks involved in sharing your images online and how people can use AI to analyze the data in those images, whether embedded or just visual, to figure out where it was taken. Get the morning's top stories in your inbox each day with our Tech Today newsletter.

ChatGPT is now a potent tool for finding the locations of photos, raising doxxing concerns

Some are concerned about the privacy implications and the potential for doxing. With the release of its latest models earlier in the week, OpenAI seems to have inadvertently tuned ChatGPT to become a potent geo-guesser. The newly available o3 and o4-mini are so good at this 'reverse location search' task that showing off this newfound functionality has become a viral social media trend, notes TechCrunch. However, this apparent geographic needle-in-a-haystack hunting improvement raises privacy concerns. And pro geo-guessers on social media platforms might be a little worried too. This newfound ability of ChatGPT is a great example of the strengthened visual reasoning being brought to the platform with model updates. It can now reason based on the content of uploaded images and perform some Photoshop-esque tasks like cropping, rotating, and zooming in. As per the source report, there are plenty of examples of users of this famous AI chatbot now using it to drill down on the location of various images. A popular jape is to ask ChatGPT to imagine it is playing the online GeoGuessr game and provide the answer based on supplied imagery. Below, we've embedded an example of ChatGPT's location divining skills, shared by AI enthusiast YouTuber and Twitterer Brendan Jowett. As Jowett points out, the newly popular ChatGPT 'reverse image search' functionality has privacy implications, and raises particular concerns with regard to doxing. Doxing is publicly sharing someone's private information, particularly location / residence, on the broad internet. People are commonly doxed with malicious intent, with the perpetrator hoping to direct loonies and cranks to visit upon the victim(s). Interestingly, TechCrunch notes that 'Geoguessr' ability isn't new for ChatGPT with the release of o3 and o4-mini. It is just the trend / awareness that has ballooned. It is said that o3 is particularly good at reverse location search, but GPT-4o, a model released without image-reasoning, can sometimes outpace o3, and deliver the same correct answer "more often than not," says TechCrunch. Before we go, it is worth mentioning that AI geo-guessing isn't a totally dependable or 100% accurate function of ChatGPT with the latest models. That's not surprising. Also, TechCrunch got a statement from OpenAI on its viral GeoGuessr success. In brief, the artificial intelligence pioneer said that, while it works to improve its tools with things like visual reasoning, it also spends time training models to refuse requests for private or sensitive information. Creative users may be able to sidestep safeguards for a time, but OpenAI indicated it will take action where it sees evidence of abuse of its usage policies.

ChatGPT gets scarily good at guessing photo locations, sparking doxxing concerns

A hot potato: Now that people have mostly stopped using ChatGPT to turn themselves into action figures, it seems the next trend involving the AI is using it to guess locations based on photos. While some are finding this reverse location search functionality fun, it raises several privacy concerns, especially when it comes to doxxing. OpenAI released its latest o3 and o4-mini models last week, which can "reason" through uploaded images. This means it can crop, rotate, and zoom in on photos, even if they're of poor quality. Combined with the models' other abilities, people have found that they are particularly good at identifying locations in uploaded photos. Users are feeding o3 images of everything from restaurant menus to selfies and telling the model to imagine it is playing the online guessing game GeoGuessr, which tasks players with guessing locations based on Google Street View images. It's easy to see this as all fun and games, but there's a potentially darker side. This reverse image search could easily allow someone to be doxxed - the public revealing of where they live or are located - based on minute details in an image that most humans would not notice. A simple selfie with few background items, or a story on social media, could be fed into ChatGPT to learn where it was taken. While users have praised the o3 model's ability to identify locations from images, it isn't something that arrived with the latest releases. TechCrunch notes that GPT-4o, which was released without image reasoning, was able to come up with the same answers as o3 more often than not, and it did so in less time. However, there was one instance in the publication's testing where o3 was able to correctly guess that a picture of a purple rhino head mounted in a bar was from a Williamsburg speakeasy - GPT-4o thought it was from a UK pub. It's important to note that even o3 doesn't get its guesses right every time, and sometimes it gets stuck in a loop when trying to determine a location. An OpenAI spokesperson said that visual reasoning will make its tools more helpful in areas like accessibility, research, or identifying locations in emergency response. As for preventing doxxing, the spokesperson said the models refuse requests for private or sensitive information, and the company has added safeguards intended to prohibit the models from identifying private individuals in images.

There's no need to overshare on social media now that OpenAI's new chatbots can pinpoint your location from the tiniest details in images

The ultimate GeoGuessr cheat code or just a privacy nightmare? Word to the wise, be careful about the images you post on social media. OpenAI's latest AI models, released last week, have sparked a new viral craze for bot-powered geoguessing. In other words, using AI to deduce where a photo was taken. Not to put too fine a point on it, but that could be a doxxing and privacy nightmare. OpenAI's new o3 and o4-mini models are both capable of image "reasoning". In broad terms, that means comprehensive image analysis skills. The models can crop and manipulate images, zoom in, read text, the works. Add to that agentic web search abilities, and you theoretically have a killer image-location tool, foreboding pun somewhat intended. According to OpenAI itself, "for the first time, these models can integrate images directly into their chain of thought. They don't just see an image -- they think with it. This unlocks a new class of problem-solving that blends visual and textual reasoning." That's exactly what early users of the o3 model in particular have found (via TechCrunch). Numerous posts are popping up across social media showing users challenging the new ChatGPT models to play GeoGuessr with uploaded images. A close-cropped snap of a few books on a shelf? The library in question at the University of Melbourne correctly identified. Yikes. Another X post shows the model spotting cars with steering wheels on the left but also driving on the left-hand side of the road, narrowing down the options to a few countries where driving on the left is required but lefthand drive cars are common, including the eventual correct guess of Suriname in South America. The models are also capable of laying out their full reasoning, including the clues they spotted and how they were interpreted. That said, research published earlier this year suggests that the explanations these models give for how they arrive at answers doesn't always reflect the AI's actual cognitive processes, if that's what they can be called. When researchers at Anthropic "traced" the internal steps used by its own Claude model to complete math tasks, they found stark differences with the method the model claimed it had used when queried. Whatever, the privacy concerns are clear enough. Simply point ChatGPT at someone's social media feed and ask it to triangulate a location. Heck, it's not hard to imagine that a prolific social media user's posts might be enough to allow an AI model to accurately predict future movements and locations. All told, it's yet another reason to be circumspect about exactly how much you spam on social media, especially when it comes to fully public posts. On that very note, TechCrunch queried OpenAI on that very concern. "OpenAI o3 and o4-mini bring visual reasoning to ChatGPT, making it more helpful in areas like accessibility, research, or identifying locations in emergency response. We've worked to train our models to refuse requests for private or sensitive information, added safeguards intended to prohibit the model from identifying private individuals in images, and actively monitor for and take action against abuse of our usage policies on privacy," was the response, which at least shows the AI outfit is aware of the problem, even if it's yet to be demonstrated that these new models would refuse to provide geolocations for any given image or collection of images.