Can ChatGPT Identify a Person in a Photo?
Published:
It’s designed not to: OpenAI’s system card for its first vision model says the model refuses when you show it a face and ask “who is this?”, and OpenAI’s usage policies forbid using it to build facial recognition databases without consent. It still sees the whole picture, can describe the scene and read any text in it, which is why there are really two questions here: what the model will say about the person, and where the photo goes once you’ve hit upload.
The model’s answer is the easy part. What happens to the file is the part to watch.
What ChatGPT refuses, and where that refusal stops
The clearest statement comes from OpenAI’s system card for GPT-4 with vision, published in September 2023. It says the model refuses requests about identity, giving two examples: a user uploads an image of a person and asks who they are, or uploads a pair of images and asks if they’re the same person. The same card lists refusals for sensitive traits such as age and race, and for conclusions drawn from traits the picture doesn’t show.
That card is three years old and today’s models are newer, so read it as the design intent, not a guarantee. Three limits apply.
Famous faces. OpenAI itself called this an open question in that card: “should models carry out identification of public figures such as Alan Turing from their images?” Its Model Spec, the rulebook for how the models should behave, lets the assistant share public, low-harm information about public figures but not private details about anyone. So don’t rely on either answer.
People you name yourself. If you type “this is my sister Marta”, ChatGPT can use that name, because you supplied it. That isn’t face recognition. It’s you adding personal data about Marta to the chat, and it gets stored with the photo.
Hidden data in the file. OpenAI’s Image Inputs FAQ says the model “doesn’t process original file names or metadata”. By OpenAI’s account, then, the model isn’t reading GPS coordinates out of the picture. They are still inside the file you uploaded, though, and the photo safety guide covers how to strip location on iPhone and Android.
What OpenAI’s rules say
OpenAI’s usage policies are blunter than any system card. Under “Respect privacy” they list what’s off limits. You may never use OpenAI’s services for “facial recognition databases without data subject consent”, for real-time remote biometric identification in public spaces, or to classify people by their biometric data. It also bans attempts to “aggregate, monitor, profile, or distribute” someone’s private information without their authorization.
Break those rules, or work around them, and the policies say you may lose access.
| What you ask about a photo with a face | Verdict |
|---|---|
| ”What’s written on this sign behind me?” | ● Safe to ask, though the photo is stored like any upload |
| ”Describe this outfit” or “is the lighting good?” | ▲ Caution: fine to ask, but crop the face if it isn’t the point |
| ”Who is this person?” | ⬣ Risk: built to refuse, and the stranger’s photo is stored anyway |
Where the photo goes after you upload it
This is the part people skip. A refusal decides what ChatGPT says. It doesn’t stop the picture being kept.
- Library. OpenAI’s Library page says ChatGPT automatically saves files you upload in chats, images included, so you can find them later. To remove one, you have to delete it from Library: deleting the chat does not delete the Library copy.
- Training. On Free, Plus and Pro, OpenAI’s data controls page says data sharing is on by default, so your uploads can help train OpenAI’s models. On Go, check the same Data controls page before you upload. Business, Enterprise and Edu workspaces are excluded by default.
- The switch. Settings, then Data controls, then turn off Improve the model for everyone. OpenAI’s data controls page says your new conversations then won’t be used for training. Chats you already had stay in your history.
- Temporary Chat. It isn’t used for training while it stays temporary, and OpenAI’s Library page says files uploaded in one aren’t saved to your Library. Its Temporary Chat page adds one exception: if you later save the chat, eligible files can be saved to Library too. OpenAI may still keep a copy of the chat for up to 30 days for safety.
None of that changes because the photo shows a face. It just matters more when the face isn’t yours.
If the photo shows someone else
Here the question stops being about ChatGPT and starts being about you. OpenAI’s Terms of Use, in both the standard and the European version, make you responsible for what you upload, and you “represent and warrant” that you have “all rights, licenses, and permissions” needed to provide it. You keep ownership of what you upload, as the ownership section of the photos guide explains, but owning a picture of your colleague doesn’t mean your colleague agreed to it going into a chatbot.
A recognizable face is that person’s personal data. In Europe that brings the GDPR into play, and if the photo comes from work, a client or a school, ask your data protection officer before uploading rather than after. For text you’re about to send with the picture, such as a caption, a name or an address, the Paste Checker flags details like an address, a phone number or an email on your own device; it spots names only roughly. It reads text only, not the image.
”I already uploaded a photo of someone”
You can undo most of this.
- Delete the chat. OpenAI says deleted chats are scheduled for permanent deletion within 30 days, unless it must keep them longer for security or legal reasons.
- Delete the Library file. Open Library, select the image, delete it. If you see Recently deleted, you can empty it there too. OpenAI says deleted files are scheduled for permanent deletion within 30 days, unless it must keep them longer for security or legal reasons.
- Turn training off for next time, as above.
- Tell the person. It’s an awkward message and the right one. “I put a photo of you into ChatGPT, I’ve deleted it, sorry” costs less than them finding out another way.
Quick answers
Will ChatGPT tell me who a stranger in a photo is?
It’s not supposed to. OpenAI’s system card for its first vision model says the model refuses when someone uploads an image of a person and asks who they are, or asks whether two images show the same person. Refusals aren’t perfect, so the safer habit is not to ask.
Can ChatGPT read where a photo was taken from the file?
OpenAI’s Image Inputs FAQ says the model doesn’t process original file names or metadata. That covers what the model reads, not what you send: the location your phone stored is still inside the file you upload, so strip it first if it matters.
I uploaded a photo of a friend. How do I remove it?
Delete the chat, then open Library and delete the image there too, because deleting a chat doesn’t delete files saved to Library. OpenAI says deleted files are scheduled for permanent deletion within 30 days, unless it must keep them longer for security or legal reasons.
If faces are only one part of what worries you, the full list of things to keep out of ChatGPT covers the rest.
Keep reading
Frequent questions
Will ChatGPT tell me who a stranger in a photo is? +
It's not supposed to. OpenAI's system card for its first vision model says the model refuses when someone uploads an image of a person and asks who they are, or asks whether two images show the same person. Refusals aren't perfect, so the safer habit is not to ask.
Can ChatGPT read where a photo was taken from the file? +
OpenAI's Image Inputs FAQ says the model doesn't process original file names or metadata. That covers what the model reads, not what you send: the location your phone stored is still inside the file you upload, so strip it first if it matters.
I uploaded a photo of a friend. How do I remove it? +
Delete the chat, then open Library and delete the image there too, because deleting a chat doesn't delete files saved to Library. OpenAI says deleted files are scheduled for permanent deletion within 30 days, unless it must keep them longer for security or legal reasons.