A user recently uploaded a screenshot to ChatGPT and found the model referencing the exact file number on their laptop - a number they say they never mentioned. The reaction was immediate: this looks like ChatGPT has access to information it shouldn't.
The most likely explanation is much less alarming: the filename.
On most operating systems, screenshots are auto-named with sequential numbers or timestamps. Windows saves them as "Screenshot (47).png" or similar. macOS names them "Screen Shot 2026-06-02 at 3.14.47 PM.png." When you upload that file to ChatGPT, the filename travels with it. ChatGPT reads that filename. If the model says "I can see this is screenshot 47" or similar, it's reading the file metadata that came with the upload - not monitoring your hard drive independently.
The ChatGPT desktop app does add a layer of nuance here. The desktop application requests permissions to access certain system information when you install it, and it can interact with files you open or share through it. Depending on which permissions you approved at setup, the app may have access to filenames, window titles, and other desktop context beyond what you explicitly paste into the chat window. OpenAI's privacy documentation states that uploaded files - including metadata - are processed by the model and can inform responses within a session.
OpenAI has not commented on this specific incident.
The practical check before assuming something unusual: look at your screenshot filename. If it contains a number, that number is visible to any system that receives the file - including ChatGPT. If you're using the ChatGPT desktop app, it's worth reviewing which permissions you granted at installation, particularly if you use it for work that involves sensitive files. You can audit and revoke those permissions through your operating system's privacy settings without uninstalling the app.