Your ’80s style AI photo is temporary: What happens to images you upload? - The Indian Express

Nostalgia seems to be inundating social media lately. The ’80s-style AI photos is a trend that is fast catching on, but this is not a new phenomenon; remember ‘Ghiblify’? These trends ask users to take the help of artificial intelligence (AI) tools to transform their images into the trending style. A younger version of yourself perched stylishly on a cherry-red Contessa is certainly an alluring thought.

It only takes a few seconds to transform your image, and it is fun, and millions are already doing it. But there is a brewing concern that many seem to overlook. A recent report in Wired sheds light on the perils of impulsively uploading images to AI chatbots.

For many users, it may just be a picture from an album on their phones, along with hundreds of others. In simple words, when you upload a selfie to ChatGPT, Gemini or any other AI app, you are offering more than a picture. It is data. Photos carry hidden metadata, which is essential information like when it was taken, what device was used, and the location. Since a face is a unique attribute, the image counts as personal data the moment it identifies who the user is.

In the report, cybersecurity expert Christina Popper of NYU Abu Dhabi revealed that AI systems collect multiple layers of information all at once. These are essentially what you type and upload and what the system infers about the user from it: background details about sessions like IP address and device type, and responses or files that get stored afterwards. So, in simple words, when you upload an image for a trend, you are giving a lot more data.

What is important to note is that the AI-generated image may vanish from the social feeds within days, but the original photo uploaded to make it could stick around for far longer than one could imagine. It depends on the data retention rules the AI company has set. According to the report, ChatGPT keeps chats until the user deletes them, but they can remain on OpenAI’s servers for up to 30 days and longer if de-identified from the account. On the other hand, Gemini retains chat data for 18 months by default. Claude removes deleted chats from Anthropic’s back-end within 30 days. However, if the user allowed their chats to help train the model, de-identified copies could be kept for up to five years.

None of the above necessarily means that an ’80s makeover photo is likely to be leaked or misused. It simply means that another copy of the user’s face now exists on an AI company’s servers under that company’s rules.

In the long run, though, the report indicates that casual use of these tools could quietly create a profile of the user. This means even without their name attached, patterns in what they uploaded and asked about can reveal their habits, interests, or profession. Now that AI tools are becoming integral to daily life with access to email, shopping, and albums, the data sitting around them becomes more valuable and likely more exposed.

The practical advice is to skip trends that ask for identity documents or overly personal images. Refrain from uploading photos of other people, especially children, without their consent. Turn off model training in the settings where that option exists. Try using temporary or incognito chat modes for anything sensitive. Delete old chats and files that are no longer needed.

AI chatbots aren’t necessarily more dangerous than other apps used every day; they are simply not private.