AI Photo Trends: What Happens to Your Photos?

The latest AI photo trend is taking social media by storm, but there’s more to these viral images than meets the eye.

The latest AI image trend buzzing across social media feeds is the ’80s AI photo. Thousands of users are casually uploading selfies to turn themselves into retro-style portraits, in a trend reminiscent of ‘Ghiblify,’ which took off in early 2025 after OpenAI rolled out upgraded image-generation features for GPT-4o.

AI chatbots can transform an image within minutes, producing results that are often strikingly realistic. But amid the fascination, few users stop to ask a more important question: where does that image go once it is uploaded? And that, precisely, is the problem.

A report in Wired sheds light on what happens behind the scenes when a user uploads their image to an AI chatbot. Let’s unpack.

A photo is more than pixels

A picture is not just pixels. According to the report, an uploaded image comes loaded with metadata, which is information about the time it was taken, the device that took it, and even the location. Since a face is identifiable, the user’s selfie counts as personal data the moment an AI system processes it.

According to cybersecurity expert Christina Popper of NYU Abu Dhabi, AI chatbots gather information in four layers – the content you upload directly, whatever the system infers about you from it, metadata tied to your session, and the outputs that get stored later. A single picture can provide all four at once.

Once the trend fades, an AI-generated portrait gently disappears from the feeds. But the original photo that a user may have uploaded to the app to create the makeover does not necessarily follow the same timeline. The image stays on the company’s servers under the company’s own retention rules.

These rules vary widely. For instance, ChatGPT retains user chats until deleted, and even then they can remain on OpenAI’s systems for up to 30 days or longer once ‘de-identified’. Meanwhile, Google’s Gemini reportedly keeps data for 18 months by default. Anthropic’s Claude clears deleted chats from the company’s back-end within 30 days, but data from chats a user has allowed for training can be kept, de-identified, for as long as five years.

All of these do not necessarily mean a user’s photo will be misused, leaked or used for deepfakes. Simply put, a copy of the user’s face now stays somewhere beyond their control.

Popper told Wired that the risk is not limited to a single photo. According to her, repeated use of these tools can surprisingly build a detailed picture of the user over time, such as their habits, interests, and even profession. And all this is inferred from what a user uploads and asks, even when their name is not attached. With AI becoming increasingly integrated into everyday life, the value of this accumulated data grows manifold, as does the potential misuse.

Practice caution

Perhaps the best way to steer clear of handing over personal data is to skip trends that seek ID documents or personal images. Avoid uploading photos of other people, especially children, without consent. Turn off model training in the account settings if it’s available. Use incognito or temporary chat modes for anything personal or sensitive. It is advisable to delete chats and files once the work is done.

The bottom line is, none of this means AI chatbots are uniquely dangerous. They are convenient, but they are just not private. So, the next viral trend may be tempting, but one should be cognisant of what they are trading for it.