Did Photos Uploaded to ChatGPT Leak Online? What Really Happened With OpenAI’s Privacy Incident
ChatGPT did not expose every photo uploaded by users. Here is what actually happened with searchable shared chats, what OpenAI changed, and how to protect your images and personal data.
Did Photos Uploaded to ChatGPT Leak Online? What Actually Happened?
People now upload photos to ChatGPT for all kinds of reasons.
They may want an image analyzed, a document read, a product photo edited, or information extracted from a screenshot.
That makes headlines claiming that “ChatGPT photos appeared on the internet” understandably worrying.
But there is an important distinction to make.
Based on the publicly available evidence, there is no basis for saying that every photo uploaded to a normal private ChatGPT conversation was exposed on the public internet.
The best-documented incident involved shared ChatGPT conversations that some users explicitly allowed to become discoverable by search engines in 2025. Those public pages could then appear in services such as Google Search.
What actually happened?
In 2025, ChatGPT users could create links to share conversations with other people.
During a short-lived experiment, users who shared a conversation could also enable an option called “Make this chat discoverable,” allowing the shared page to appear in web searches.
On July 31, 2025, TechCrunch reported that ChatGPT shared pages could be found through Google and other search engines.
OpenAI later said the search-discoverability feature had been an experiment and removed it after concerns that users could accidentally make information public that they did not intend to share.
Category | Details |
|---|---|
Issue became widely reported | Late July 2025 |
Service involved | ChatGPT shared conversations |
Type of issue | Search-engine indexing of shared pages |
Data potentially visible | Content included in the shared conversation |
Users affected | Certain users who shared chats and enabled discoverability |
Confirmed server breach | Not established by the available reporting |
OpenAI response | Removed the discoverability feature and worked to de-index pages |
Current status | Shared-link pages are not intended for search-engine indexing |
Were photos uploaded to ChatGPT really published online?
Not automatically.
Uploading a photo to a private ChatGPT conversation and publishing information on the public web are two different things.
There is no evidence from this incident that merely uploading an image to a normal private ChatGPT conversation automatically made it searchable on Google.
However, shared conversations are different.
OpenAI's current documentation says a shared conversation may include conversation history, supported images, or uploaded files, depending on the sharing experience and permissions.
That means an image included in a conversation could potentially become visible if that conversation itself was deliberately shared.
A useful way to think about it is:
Private upload: not automatically a public web page
Shared conversation: accessible to people who can open the shared link
2025 discoverability option: could make certain shared pages searchable
AI-generated image: separate from an uploaded personal photo and not automatically public
So describing the event as an “OpenAI image leak” can be misleading if it suggests that OpenAI's entire private image database was exposed.
Were all ChatGPT users affected?
No evidence shows that all ChatGPT users were affected by this issue.
The search indexing depended on specific sharing behavior.
Users first had to create a shared conversation, and the experimental search discoverability option required an additional choice. OpenAI described the feature as opt-in.
Reports found thousands of indexed shared conversations, but that number should not be interpreted as evidence that every user's private conversations or images were leaked.
The exact number of users whose personally identifiable images were viewed, copied, or retained by third parties has not been publicly established.
Was OpenAI hacked?
The widely reported 2025 incident was not described as an attacker breaking into OpenAI's servers and stealing a private database.
It was primarily a visibility and sharing problem.
That distinction matters.
A data breach caused by an attacker bypassing security controls is different from a webpage being publicly accessible because of a sharing setting.
Both can create privacy risks, but they have different causes.
Security researchers discussing the incident similarly distinguished it from a traditional external compromise.
Why could shared conversations appear in search engines?
Search engines such as Google and Bing automatically discover publicly accessible webpages.
The process of collecting pages is generally called crawling.
Adding information about those pages to a search database is called indexing.
During the 2025 experiment, certain ChatGPT shared pages could be made discoverable to search engines. If a search engine crawled and indexed such a page, it could appear in search results.
This did not mean that Google was freely reading every user's private ChatGPT account.
The pages at issue had been made available through ChatGPT's sharing mechanism.
What did OpenAI do?
OpenAI removed the search-discoverability feature on July 31, 2025.
The company said the feature created too many opportunities for people to accidentally share content they had not intended to make searchable.
OpenAI also said it was working to remove previously indexed material from relevant search engines.
As of September 2026, OpenAI's Help Center states that shared-link pages are not intended for search-engine indexing.
However, a shared link should still not be treated as a private vault.
For personal accounts, anyone who has access to the link may be able to view the shared content and forward the link to someone else.
Is the same problem still happening today?
The specific discoverability feature that triggered the 2025 controversy has been removed.
It would therefore be misleading to describe that old feature as if it were still operating unchanged today.
The more relevant current privacy concern is ordinary sharing.
A shared conversation can expose information to people who receive or obtain the link, and supported images or files may be included in the shared content.
In other words, the practical question today is less “Will Google automatically publish my photo?” and more “Am I sharing a conversation that contains information I would rather keep private?”
What should you check before uploading a photo?
Photos often contain more personal information than people realize.
A single image can show a face, vehicle registration number, school name, workplace badge, home interior, mailing label, medical document, or other identifying details.
South Korea's Personal Information Protection Commission has recently advised users of generative AI services to avoid entering highly sensitive information such as passwords and identification numbers, avoid uploading confidential workplace documents without appropriate safeguards, and review AI training and data-retention settings.
Information | Safer approach |
|---|---|
Face photos | Upload only when necessary |
Children's photos | Check for school, name and location clues |
Passport or ID | Mask ID numbers, address and date of birth |
Financial records | Remove account and card numbers |
Company documents | Follow workplace AI and security policies |
Medical or legal documents | Remove unnecessary identifying information |
Vehicle plates | Blur them when they are not relevant |
Shared links | Review everything included before sharing |
Model training | Review “Improve the model for everyone” under Data controls |
OpenAI currently allows eligible users to turn off “Improve the model for everyone.” When disabled, new conversations are not used to improve OpenAI models. This setting does not itself delete existing chats.
What should I do if I already uploaded photos?
There is no need to panic and delete every ordinary image simply because you once uploaded it to ChatGPT.
However, it is worth reviewing conversations that contain especially sensitive information, such as:
passports or national IDs
credit cards or bank details
government identification numbers
medical records
legal contracts
confidential workplace information
children's school or location information
clearly visible home addresses
If you no longer need a chat, you can delete it.
OpenAI states that deleted chats disappear from the account immediately and are generally scheduled for permanent deletion from its systems within 30 days, subject to limited legal, security, or de-identification exceptions.
There is one additional detail worth knowing in 2026.
When Library is available, uploaded files may be stored separately from the conversation. Deleting the chat therefore may not delete a file that remains saved in Library. Users who want the file removed should check Library as well.
If you have previously created shared links, review them under:
Settings → Data controls → Shared links → Manage
OpenAI currently lets users delete individual shared links or delete all shared links.
Deleting the original conversation also removes its shared-conversation link. However, copies already saved by another user are not automatically removed from that user's account.
The bottom line
The claim that “all photos uploaded to ChatGPT leaked onto the internet” overstates what the evidence shows.
The major 2025 incident involved an experimental feature that allowed some deliberately shared ChatGPT conversations to become discoverable through search engines.
Because shared conversations could contain personal information and, in some circumstances, supported images or uploaded files, the incident created a genuine privacy concern.
But it was not established as a mass hack exposing every private ChatGPT user's uploaded images.
The search-discoverability feature was removed.
The broader privacy lesson remains useful: upload only the information you actually need, remove unnecessary identifying details, check sharing settings carefully, and avoid putting highly sensitive personal or confidential information into an AI service unless you understand how it will be handled.
A simple rule is helpful:
Before uploading an image, ask yourself whether the image contains anything you would strongly regret exposing to someone else.
If the answer is yes, mask or remove that information first.
Frequently Asked Questions
Q1. Are all photos uploaded to ChatGPT publicly available online?
No. There is no evidence from the 2025 incident that normal private image uploads were automatically published on the public web.
Q2. Should I delete every photo I have previously uploaded?
Not necessarily. Focus on especially sensitive material such as identity documents, financial information, medical records, children's location information and confidential workplace material.
Q3. Did hackers break into OpenAI and steal users' photos?
The 2025 search-indexing incident was not established as a server intrusion or database theft. It involved shared pages and search-engine discoverability.
Q4. Is it safe to upload photos of children?
Use extra caution. Check whether the image reveals names, school information, uniforms, addresses, landmarks or other clues that could identify the child's location.
Q5. Can shared ChatGPT links still appear in search engines?
OpenAI currently says shared-link pages are not intended for search-engine indexing. However, anyone who can access a personal-account shared link may still be able to view and forward its contents.