Is it safe to upload photos? Privacy risks, metadata, and data leakage (ChatGPT caricature trend)

Lisa Ernst · 04.02.2026 · Technology · 6 min

I treat every “quick upload” as a potential data leak until I’ve inspected the image the same way I’d review a document before sending it outside my organization.

Source: Own illustration

Is it safe to upload my photo?

It can be safe enough for many people, but it’s rarely “risk-free.” The obvious fear is “someone steals my face,” but the more common problem is what the photo quietly reveals: where it was taken, who else appears in it, what’s on a badge, what sits in the background, and what the service may retain or reuse.

For OpenAI’s consumer services, OpenAI explains that content you provide may be used to improve services (including model training), with opt-out options via data controls and the privacy portal in How your data is used to improve model performance. The broader collection and handling of personal data is described in the Privacy Policy.

If you use Temporary Chat, OpenAI says those chats won’t appear in history and won’t be used to improve models; they may be retained for up to 30 days for safety purposes (see the Temporary Chat FAQ). Separately, OpenAI states that deleted chats are scheduled for permanent deletion within 30 days, with listed exceptions, in its Chat and File Retention Policies.

One practical gotcha: if you use a custom GPT with “actions,” data can be sent to third parties. OpenAI notes that this data is then subject to the recipient’s privacy policy and may be retained longer than 30 days (also covered in the Temporary Chat FAQ).

What a “simple selfie” can expose

1) Location, time, and device details (metadata)

Many photos include metadata (often called EXIF): time and date, camera/device details, and sometimes GPS location. Apple explains how location metadata can travel with shared photos in its guide on managing location metadata, and Canon provides a practical overview in All about EXIF. Great for organizing your library — risky when you share the original file.

2) Other people’s faces (bystanders, patients, colleagues, children)

If someone is recognizable, you’re not only uploading your information. The Swiss Federal Data Protection and Information Commissioner notes that photos are personal data as soon as someone can be identified — including group photos — in its guidance on photos and privacy.

3) Badges, screens, and “workplace tells”

The easiest leaks are the boring ones: an ID badge with your full name, a staff-only door sign, a patient label on a clipboard, a whiteboard roster, or even a monitor reflection. No hacking required — just zoom and context.

4) Biometric sensitivity (when faces are used for identification)

Under EU rules, biometric data processed for uniquely identifying a person is treated as sensitive personal data. The European Commission summarizes this in its explanation of what personal data is considered sensitive. This matters whenever an image can be linked back to you in a way that enables recognition or profiling.

Photo privacy checklist for the ChatGPT caricature trend

1. Decide your “acceptable exposure” first. If you’d be uncomfortable seeing this image on a shared screen at work, don’t upload it. (That’s a judgment call, not a technical step.)
2. Crop aggressively. Keep only what the model needs — usually head + hairline. Everything else increases the risk surface.
3. Remove bystanders completely. If you can’t crop them out, don’t upload the photo. Recognizable people can make the image personal data about them (see the FDPIC guidance on photos and privacy).
4. Blur or cover IDs and uniforms. Especially: name badges, access cards, patient wristbands, school logos, vehicle plates. If it links you to a place, it can link you to a routine.
5. Use a neutral background. A blank wall beats a kitchen counter (mail), an office wall (org charts), a ward corridor (unit signage), or a classroom (children).
6. Strip location metadata before you upload. Apple shows how to manage and remove location metadata in this guide. For Google Photos, location sharing behavior depends on your settings, as explained in Google Photos’ location information help page.
7. Prefer Temporary Chat for one-off image fun. OpenAI says Temporary Chats won’t be used to improve models and may be retained up to 30 days for safety (see the Temporary Chat FAQ).
8. Avoid custom GPT “actions” if you don’t want third-party data sharing. OpenAI notes that when actions send data to third parties, it becomes subject to the recipient’s privacy policy (see the Temporary Chat FAQ).
9. Control training/usage settings. OpenAI outlines opt-out options (data controls + privacy portal) in How your data is used, and the broader data handling is described in the Privacy Policy.
10. Use a “no-face” alternative when possible. If your goal is the caricature style — not your exact likeness — use a text prompt (“a friendly caricature of a person with curly hair and round glasses”) or upload an image without a face (hands, outfit, silhouette). This removes biometric exposure entirely in the sense described under EU sensitive-data context (see what personal data is considered sensitive).

Fast metadata wipe (batch) with ExifTool

If you want a reliable, repeatable way to remove metadata, ExifTool is a widely used utility for reading, writing, and deleting metadata. The official documentation also details write/delete operations in the ExifTool POD.

exiftool -all= -overwrite_original "photo.jpg"

# Batch example (folder):
exiftool -all= -overwrite_original *.jpg

On Windows, you can also use File Explorer to remove properties from a file. The EU Publications Office documents that approach in Removing metadata and personal information.

Quick decision guide

Hands-on walkthroughs (metadata removal)