OpenAI details ChatGPT privacy safeguards and user controls

OpenAI has published a plain-language guide explaining how ChatGPT learns from data while protecting privacy. The company uses a mix of publicly available information, partnership data, and user-provided content to train its models. To reduce personal information in training datasets, OpenAI developed Privacy Filter, an internal tool that identifies and masks personal text, which it claims outperforms any comparable tool. Privacy Filter is applied at multiple stages, including on public datasets and user conversations when the 'Improve the model for everyone' setting is enabled. OpenAI has also made Privacy Filter freely available to other developers. Users can opt out of training by turning off the setting in Data Controls, or use Temporary Chat, which does not appear in history, create memories, or contribute to model improvement; conversations are retained for 30 days for safety then deleted. The Memory feature is optional and can be reviewed, edited, or turned off entirely. ChatGPT is designed to reject requests for private information, but users can submit privacy requests via the portal if errors occur. OpenAI emphasizes its commitment to balancing privacy with safety, including detecting credible threats of violence.