OpenAI employs hundreds of contract workers via Mercor to review anonymized ChatGPT conversations for model improvement, with reviewers earning over $50 hourly. This practice, aimed at reducing issues like excessive flattery, raises privacy concerns as users may not realize their data is manually reviewed despite privacy filters. OpenAI defaults to using chat data for training but offers an opt-out setting. Similar human review mechanisms exist at Anthropic and Google, highlighting the ongoing tension between AI training efficiency and user data privacy.