Does anyone else have this? Inspector tool shows data_usage_for_training: “permitted” but the toggle is set to opt-out (OFF)

Understanding Data Usage Settings and Potential Discrepancies in ChatGPT

In the landscape of AI and data privacy, transparency and user control over personal data are paramount. Recently, I conducted an investigation into the data usage settings within ChatGPT and encountered a potential discrepancy that warrants attention from users and developers alike.

The Discovery

While exploring the ChatGPT interface out of curiosity, I navigated to Settings > Data Controls and observed the toggle labeled “Improve the model for everyone.” I had personally set this toggle to OFF over a year ago, aiming to prevent my data from being used for model training.

However, upon inspecting the network responses via the browser’s Developer Tools, I noticed an unexpected detail. A specific JSON response included a field:

data_usage_for_training: "permitted"

Despite the toggle being set to OFF, the system seemed to indicate that data usage for training was permitted.

This inconsistency raised concerns about whether my data was indeed being excluded from model training or if there was a bug in how settings are reflected in system responses.

Steps to Investigate

To verify this behavior, I followed these steps:

1. Loaded the ChatGPT webpage in my browser.
2. Opened Developer Tools (F12 or right-click > Inspect).
3. Navigated to the Network tab.
4. Accessed Settings and clicked on Data Controls.
5. Filtered network requests to locate the response containing data_usage_for_training.
6. Examined the response payload, focusing on the specific fields such as auth_user_id (confidential) and data_usage_for_training.

Observations

• Changing the toggle from OFF to ON or vice versa did not alter the value of data_usage_for_training; it remained “permitted.”
• Refreshing the page after toggling did not update the status in network responses.
• This suggests that the setting’s visual toggle may not directly influence the underlying data usage setting as reflected in system responses, indicating a possible bug or a disconnect between UI and backend state.

Implications

If this behavior is consistent across user accounts, it raises significant privacy concerns. Users aiming to opt out of data training may believe their preferences are respected, but system responses could be conflicting, leading to unintended data sharing.

Next Steps

I recommend the following:

• Users should remain cautious and verify their actual data sharing status through multiple means.
• Developers should review the synchronization between UI toggles and backend data flags.
• A transparent explanation from the platform about how data usage