OpenAI uses hundreds of contractors to review ChatGPT chats

OpenAI’s disclosure of this practice is limited to a buried FAQ entry that has remained largely unchanged since 2023. The company argues that explicit consent for human review would likely result in high opt-out rates, undermining the human feedback loop essential for model improvement. Users can disable the "Improve the model for everyone" setting, which is enabled by default, or use temporary chat mode to prevent data usage for training. However, this setting only applies to new conversations and does not guarantee that existing data is never accessed by human reviewers.

This practice is not unique to OpenAI. Anthropic confirmed it uses human reviewers for Claude, and Google notes that humans may review saved Gemini chats. The widespread use of human-in-the-loop evaluation highlights a significant tension between model improvement strategies and user privacy expectations in the AI industry.