AI Briefing
KO

How ChatGPT Learns About the World While Protecting Personal Information

·2026.05.06 17:00

Key point

OpenAI has disclosed the data used to train ChatGPT and the privacy safeguards in place.

Details

ChatGPT is getting stronger at code, research, analysis, and multi-step tasks, and OpenAI explains that these capabilities come from training on diverse data. At the same time, the company disclosed the safeguards and user controls designed to minimize personal information as much as possible during the training process.

Training data consists of publicly available internet information, information accessed through partnerships, and information provided or generated by users, contractors, and researchers. For public content, only freely accessible material is used, and materials such as public forum posts or blogs can also be subject to training.

To reduce personal information, OpenAI applies the OpenAI Privacy Filter at multiple stages. This tool detects and masks personal information in text, and is used both on public datasets and on user conversations where Improve the model for everyone is turned on. OpenAI stated that, based on its own evaluations, this tool is more effective than comparable tools, and it has also been made freely available via Hugging Face.

Users can decide for themselves whether ChatGPT uses their conversations to train future models.

  • Turning off Improve the model for everyone in Settings > Data Controls means new conversations remain in history but are not used for training.
  • Temporary Chat leaves no history or memory, is not used for model improvement, and is deleted after 30 days of retention for security purposes.
  • Memory stores information needed to help with responses, but it can be reviewed, edited, deleted, or turned off completely at any time.

Users can also export their data, delete their account, and submit privacy-related requests through the privacy request portal. ChatGPT is designed to refuse requests for sensitive personal information, but errors are possible, and inaccurate or inappropriate personal information outputs can be corrected through deletion requests. OpenAI stated it will strengthen both privacy protection and its response to threats of violence.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.