OpenAI Details ChatGPT Privacy Controls and Training Safeguards
Disable the "Improve the model for everyone" setting for sensitive data, and integrate the free OpenAI Privacy Filter into your own data pipelines.
Disable the "Improve the model for everyone" setting for sensitive data, and integrate the free OpenAI Privacy Filter into your own data pipelines.
Summary
OpenAI has released a plain‑language guide explaining how ChatGPT learns while protecting user privacy. The company uses publicly available data, partnership data, and user‑generated content for training, but applies an internal Privacy Filter that masks personal information before it enters the model. The filter is now free for developers to use in their own workflows.
Users can control whether their conversations help train future models by toggling the "Improve the model for everyone" setting in Data Controls. If turned off, chats remain in history but are excluded from training. ChatGPT also offers a "Temporary" chat mode that never appears in history, does not create memories, and is automatically deleted after 30 days for safety.
Memory is optional; users can edit, delete, or turn it off entirely, preventing ChatGPT from referencing past chats. Users can export data, delete accounts, and submit privacy requests via the privacy portal if they see inaccurate or inappropriate personal information in outputs. OpenAI stresses that it will keep improving safeguards and clarifying privacy controls as models become more capable.
Key changes
- OpenAI introduces an internal Privacy Filter that masks personal information during training, available for free to developers
- Users can toggle the "Improve the model for everyone" setting in ChatGPT Data Controls to stop conversations from being used for training
- ChatGPT offers a "Temporary" chat mode that does not appear in history, does not create memories, and is deleted after 30 days
- Memory feature is optional; users can edit, delete, or turn off saved memories, preventing future references
- Users can export data, delete accounts, and submit privacy requests via the privacy request portal
- ChatGPT rejects requests for private or sensitive information but may still output personal data; users can file a privacy request if inaccurate
- OpenAI plans to continue improving safeguards, making privacy controls clearer, and giving users more control over data usage