ChatGPT's Terms Treat Your Face as "Content": Reading OpenAI's Terms on Uploaded Photos
Table of Contents
- TL;DR
- First, know which three documents you are reading
- Terms of Use: a photo is "Input", the same category as typing
- Service Terms: consent to likeness, and the subject of every sentence is "you"
- Training help page: services for individuals default to may, business contracts default to no
- The terms admit three things, and not a fourth
- Closing
- Further reading
TL;DR
- There are three documents to read, not one: the Terms of Use, the Service Terms and the training help page, and they do not carry the same legal weight.
- The Terms of Use file the images you upload under "Input", the same category as the text you type; together they are called "Content".
- Ownership is written in your name, and usage rights go to the other side at the same time. Both sentences sit in the same section.
- In the Service Terms passage on likeness, the subject is almost always "you": you may not, you represent, you give OpenAI the right.
- Opting out governs conversations from that point on.
- Nowhere in the terms does the face get a section of its own.
First, know which three documents you are reading
Most people arguing about this have not opened a single document. Once you open them, the first job is to be clear about which one you are reading.
| Document | What it is | What it decides |
|---|---|---|
| Terms of Use | The main contract, the one you click to accept | The definition of Content, ownership, what the other side may do with it |
| Service Terms | Additional terms for specific features | Visual Capabilities, likeness, the scope of the licence for public sharing |
| Training help page | An explainer that restates "improve our Services" as "train our models" | Which kinds of accounts are used for training by default |
The third is an explainer, not a substitute for the contract. It helps you understand how the first two work in practice, but what binds the two parties is still the contract.
Then there is the version question. Readers in Taiwan need the Rest of World version, the one with row-terms-of-use in its URL, effective 1 January 2026. The European Economic Area, Switzerland and the UK are covered by a separate one. Carry a discussion of the EU version over to Taiwan and the conclusions go wrong.
Terms of Use: a photo is "Input", the same category as typing
In the Terms of Use, find the section called "Content". It has five subheadings, in this order: Your content, Ownership of content, Similarity of content, Our use of content, Opt out.
The order alone already tells the whole story.
The first paragraph:
Your content. You may provide input to the Services (“Input”), and receive output from the Services based on the Input (“Output”). Input and Output are collectively “Content.” You are responsible for Content, including ensuring that it does not violate any applicable law or these Terms. You represent and warrant that you have all rights, licenses, and permissions needed to provide Input to our Services.
In plain terms: the selfie you upload has the status of "Input" in this contract, in the same category as a line of text you type, under the same rules. Not one sentence says images get special handling.
Then ownership:
Ownership of content. As between you and OpenAI, and to the extent permitted by applicable law, you (a) retain your ownership rights in Input and (b) own the Output.
In plain terms: the image is still yours, and the generated one counts as yours too. This sentence is often held up as proof that "so there is no problem", but it answers "who does this image belong to", not "what will it be used for".
That question is in the next subheading:
Our use of content. We may use Content to provide, maintain, develop, and improve our Services, comply with applicable law, enforce our terms and policies, and keep our Services safe.
In plain terms: ownership is yours, and usage rights go to the other side at the same time. There is no conflict. They are two different questions that happen to sit in adjacent paragraphs.
"Develop and improve our Services" covers more ground than most people assume. Read it together with the next subheading:
Opt out. If you do not want us to use your Content to train our models, you can opt out by following the instructions in this article. Please note that in some cases this may limit the ability of our Services to better address your specific use case.
This paragraph carries two pieces of information. First, opting out is possible. Second, it is opt out, not opt in. The default is written into the word itself.
Service Terms: consent to likeness, and the subject of every sentence is "you"
Visual Capabilities are covered in the Service Terms, not in the main contract. Verbatim:
Our models can accept images and videos as part of Inputs to the Services (“Visual Capabilities”). You may not use Visual Capabilities to assist in identifying a person nor to solicit or infer private or sensitive information about a person. You may not use Visual Capabilities to reproduce the likeness of any person without express consent and all necessary rights. By choosing to share an image or video publicly on the Services, such as by sharing it to the Sora feed or uploading it to create a Sora cameo, you represent you have all necessary rights, and you give OpenAI the right to reproduce, distribute, modify, display and perform it for the purpose of operating and promoting the Services. You also agree to give users the limited right to reproduce and remix that content solely on the Services. Uploading a cameo or sharing an image or video does not give other users any rights to use those materials outside of the Services.
Pull out the verbs and line them up: you may not (identify), nor (solicit or infer), you may not (reproduce), you represent, you give, you also agree.
Six actions, and the subject of every one is you.
This passage does protect likeness, but look carefully at which way it points. What it prohibits is users using Visual Capabilities to identify other people or reproduce other people's faces. It deals with your infringement risk toward third parties and the platform's own legal risk. It makes no promise that the platform will not extract anything from your selfie. That would be a different kind of sentence, and it is not here.
There is also a point about context that is easy to misread. The broader licence in the second half comes with a condition in front of it: "By choosing to share an image or video publicly on the Services". It is triggered by sharing publicly. Dropping an image into a private conversation does not hand it over automatically.
Ordinary uploads in a conversation fall under the "Content" section of the Terms of Use; only public sharing on Sora, or a cameo, reaches this broader licence. Treating the two as the same thing overstates the conclusion, and an overstated conclusion is easy to knock down entirely with one line: "you didn't even understand the terms".
Training help page: services for individuals default to may, business contracts default to no
The third document is called "How your data is used to improve model performance"; the Traditional Chinese edition we read was updated 13 March 2026. It turns the first two documents' "improve our Services" into plain language. The three sentences quoted below are OpenAI's earlier English wording, the text that edition translates. OpenAI has since reworded the page in both languages; what the three sentences say still holds there, but check the live page before quoting it.
When you use our services for individuals such as ChatGPT or DALL•E, we may use your content to train our models.
On the same page, the business section is more direct:
We don’t use content from our business offerings such as ChatGPT Team, ChatGPT Enterprise, and our API Platform to train our models.
Put the two sentences side by side and the difference at the account level needs no explanation from me. This is not generosity. Paying enterprise customers write this clause into their procurement contracts.
On opting out:
Once you opt out, new conversations will not be used to train our models.
The key word is "new". Opting out governs what comes after: switch it off today and it does not reach the image you uploaded last week.
As for how to opt out, this official page gives two routes: click "do not train on my content" in the privacy portal, or follow the Data Controls FAQ to turn off training for your ChatGPT conversations. The names of settings in the app change between versions, so go by what you see at the time. Part 4 quotes the opt-out wording as OpenAI's help pages read on 26 September 2026, including the exception for conversations you send feedback on, and notes that the same setting is named three different ways across the Traditional Chinese help pages.
Other providers are structured much the same way, and the defaults often differ between free and paid tiers. Look up the specific switch names on each provider's data controls page, and do not trust the names any article (this one included) passes on. Those names change faster than articles do.
The terms admit three things, and not a fourth
| What the terms admit | What the terms do not say |
|---|---|
| The image you upload is "Content", and ownership stays in your name | That faces or biometric data are a category requiring separate treatment |
| Services for individuals may use content for development and improvement, with an opt-out | Withdrawing a piece of data from a training batch that has already been completed |
| The uploader vouches for the rights of everyone in the image | That the platform will not build re-identifiable representations from selfies |
Every item in the left column is written out clearly, with nothing hidden. The right column is not hidden either: the field simply does not exist.
If the three items in the right column were written into the terms, they would become commitments someone could be held to. Left unwritten, there is no obligation and no remedy. Reading terms is not only about reading what they say. It is also about seeing which questions they leave outside the frame.
Closing
What joining that trend means in contract terms fits in one sentence: you granted a general licence over a high-resolution frontal photo of your face, declared at the same time that you had the right to do so, and vouched for every person in the frame.
The Story post packages this act as "your turn".
This is not a claim that the terms are unlawful, nor that the platform is acting in bad faith. The terms are so plain that the order of five subheadings lays the logic out. The problem is that nobody presses upload while reading those five subheadings.
What Ultra Lab does is scan applications for data being sent into models that should not go in. Here, the original terms prove only one thing: the platform has not put faces into its own definition of "should not". For now, that line is drawn by users themselves.
Read the original, not my paraphrase. Open the Terms of Use, find the "Content" section, and read its five subheadings in order. Ten minutes, and more accurate than any cheat sheet.
Further reading
- Part 1 of this series: You think you are asking ChatGPT what you would look like; the system is collecting a biometric credential
- Part 3 of this series: If your product takes photo uploads, you are already building a biometric system
- We defined the security standard for the AI era: AASS v1.0
- OpenAI Terms of Use (Rest of World), English edition; Traditional Chinese edition
FAQ
Which OpenAI Terms of Use should users in Taiwan read?
The Rest of World version, the one with row-terms-of-use in its URL, effective 1 January 2026. The European Economic Area, Switzerland and the UK are covered by a separate version. Carry a discussion of the EU version over to Taiwan and the conclusions go wrong.
What does an uploaded photo count as under the terms?
"Input". The Terms of Use state that Input and Output are collectively called "Content", so a photo sits in the same category as a line you type and follows the same rules. Nowhere do the terms single out faces or biometric data for separate treatment.
The terms say the image is still mine. Does that mean there is no problem?
Ownership and use are two separate questions. The "Ownership of content" paragraph does say you retain ownership of Input, but the adjacent "Our use of content" paragraph also says Content may be used to provide, maintain, develop and improve the Services. The two sentences sit side by side in the same section without conflict.
Does the likeness passage in the Service Terms protect me?
Look carefully at which way it points. What it prohibits is users using Visual Capabilities to identify other people or reproduce other people's faces; it deals with your infringement risk toward third parties and the platform's own legal risk. It makes no promise that the platform will not extract anything from your selfie.
After I opt out, will content I uploaded earlier be removed?
No. The official help page on training, in the edition this article quotes, says: "Once you opt out, new conversations will not be used to train our models." The key word is "new": opting out governs what comes after. Separately, that edition states explicitly that content from business offerings (ChatGPT Team, Enterprise, API Platform) is not used for training, a different default from services for individuals. OpenAI has since reworded the page, so check the live version before quoting it.